Imitation learning has shown success in many tasks by learning from expert demonstrations. However, most existing work relies on large-scale demonstrations from technical professionals and close monitoring of the training process. These are challenging for a layperson when they want to teach the agent new skills. To lower the barrier of teaching AI agents, we propose Interactive Policy Restructuring and Training (InterPReT), which takes user instructions to continually update the policy structure and optimize its parameters to fit user demonstrations. This enables end-users to interactively give instructions and demonstrations, monitor the agent's performance, and review the agent's decision-making strategies. A user study (N=34) on teaching an AI agent to drive in a racing game confirms that our approach yields more robust policies without impairing system usability, compared to a generic imitation learning baseline, when a layperson is responsible for both giving demonstrations and determining when to stop. This shows that our method is more suitable for end-users without much technical background in machine learning to train a dependable policy.

Exercise supports healthy aging, but motivation often declines with age, increasing demand on therapists and coaches. We present a conversational robotic exercise coach that promotes engagement and assesses motivation through dialogue. In a WoZ study with ten adults aged 59 and above, participants showed varied interaction styles; even those with low motivation rated sessions positively, suggesting such agents can enhance exercise enjoyment. We identify three design needs for autonomous coaches: rephrasing for clarity, conversation beyond exercise, and adaptable speech delivery.

Different people respond to feedback and guidance in different ways, and their preferences may change based on their mood, tiredness, etc. We present a robot exercise coach that provides verbal and nonverbal feedback in two different styles: firm and encouraging. We collect a dataset of people experiencing both feedback styles and show that the style that someone performs best with may not be the one they have the best subjective experience with or be the one that they state they prefer. To account for this, we present a contextual bandit approach that enables the robot coach to learn the best style to use over time to improve the human's performance, and show that this approach performs quite well in expectation on the real human data.

In interactive imitation learning (IL), uncertainty quantification offers a way for the learner (i.e. robot) to contend with distribution shifts encountered during deployment by actively seeking additional feedback from an expert (i.e. human) online. Prior works use mechanisms like ensemble disagreement or Monte Carlo dropout to quantify when black-box IL policies are uncertain; however, these approaches can lead to overconfident estimates when faced with deployment-time distribution shifts. Instead, we contend that we need uncertainty quantification algorithms that can leverage the expert human feedback received during deployment time to adapt the robot's uncertainty online. To tackle this, we draw upon online conformal prediction, a distribution-free method for constructing prediction intervals online given a stream of ground-truth labels. Human labels, however, are intermittent in the interactive IL setting. Thus, from the conformal prediction side, we introduce a novel uncertainty quantification algorithm called intermittent quantile tracking (IQT) that leverages a probabilistic model of intermittent labels, maintains asymptotic coverage guarantees, and empirically achieves desired coverage levels. From the interactive IL side, we develop ConformalDAgger, a new approach wherein the robot uses prediction intervals calibrated by IQT as a reliable measure of deployment-time uncertainty to actively query for more expert feedback. We compare ConformalDAgger to prior uncertainty-aware DAgger methods in scenarios where the distribution shift is (and isn't) present because of changes in the expert's policy. We find that in simulated and hardware deployments on a 7DOF robotic manipulator, ConformalDAgger detects high uncertainty when the expert shifts and increases the number of interventions compared to baselines, allowing the robot to more quickly learn the new behavior.

Behavior cloning has shown success in many sequential decision-making tasks by learning from expert demonstrations, yet they can be very sample inefficient and fail to generalize to unseen scenarios. One approach to these problems is to introduce general domain knowledge, such that the policy can focus on the essential features and may generalize to unseen states by applying that knowledge. Although this knowledge is easy to acquire from the experts, it is hard to be combined with learning from individual examples due to the lack of semantic structure in neural networks and the time-consuming nature of feature engineering. To enable learning from both general knowledge and specific demonstration trajectories, we use a large language model’s coding capability to instantiate a policy structure based on expert domain knowledge expressed in natural language and tune the parameters in the policy with demonstrations. We name this approach the Knowledge Informed Model (KIM) as the structure reflects the semantics of expert knowledge. In our experiments with lunar lander and car racing tasks, our approach learns to solve the tasks with as few as 5 demonstrations and is robust to action noise, outperforming the baseline model without domain knowledge. This indicates that with the help of large language models, we can incorporate domain knowledge into the structure of the policy, increasing sample efficiency for behavior cloning.

Demonstrations are a powerful way to teach robot decision-making to humans. Although informative demonstrations may be selected a priori using the machine teaching framework, student learning may deviate from the pre-selected curriculum in situ. This article thus explores augmenting a curriculum of pre-selected demonstrations with a closed-loop teaching framework inspired by principles from the education literature, such as the zone of proximal development and the testing effect. We utilize tests accordingly to close the loop and maintain a novel particle filter model of human beliefs throughout the learning process, allowing us to provide demonstrations that are targeted at the human’s current understanding in real time. A user study finds that our proposed closed-loop teaching framework reduces the regret (i.e., the suboptimality) of human test responses by 43% over an open-loop baseline. We also compare our closed-loop teaching framework against another baseline of directly communicating the robot’s reward function in a second user study. We find that our closed-loop teaching outperforms direct reward communication by 64%, but we also observe synergies from the use of both teaching forms. Finally, we observe strong interaction effects between the teaching form and the domains considered in both user studies, seeing increased learning outcomes from well-designed demonstration-based teaching in the more challenging domain.

Large language models contain noisy general knowledge of the world, yet are hard to train or fine-tune. In contrast cognitive architectures have excellent interpretability and are flexible to update but require a lot of manual work to instantiate. In this work, we combine the best of both worlds: bootstrapping a cognitive-based model with the noisy knowledge encoded in large language models. Through an embodied agent doing kitchen tasks, we show that our proposed framework yields better efficiency compared to an agent entirely based on large language models. Our experiments also indicate that the cognitive agent bootstrapped using this framework can generalize to novel environments and be scaled to complex tasks.

As robots are deployed in human spaces, it is important that they are able to coordinate their actions with the people around them. Part of such coordination involves ensuring that people have a good understanding of how a robot will act in the environment. This can be achieved through explanations of the robot’s policy. Much prior work in explainable AI and RL focuses on generating explanations for single-agent policies, but little has been explored in generating explanations for collaborative policies. In this work, we investigate how to generate multi-agent strategy explanations for human-robot collaboration. We formulate the problem using a generic multi-agent planner, show how to generate visual explanations through strategy-conditioned landmark states and generate textual explanations by giving the landmarks to an LLM. Through a user study, we find that when presented with explanations from our proposed framework, users are able to better explore the full space of strategies and collaborate more efficiently with new robot partners.

In this work, we aim to improve transparency and efficacy in human-robot collaboration by developing machine teaching algorithms suitable for groups with varied learning capabilities. While previous approaches focused on tailored approaches for teaching individuals, our method teaches teams with various compositions of diverse learners using team belief representations. We investigate various group teaching strategies, such as focusing on individual beliefs or the group’s collective beliefs, and assess their impact on learning robot policies for different team compositions. Our findings reveal that team belief strategies produce less variation in learning duration and better accommodate diverse teams compared to individual belief strategies, suggesting their suitability in mixed proficiency settings with limited resources. In contrast, individual belief strategies provide a more uniform knowledge level, particularly effective for homogeneously inexperienced groups. Our study indicates that the effectiveness of the teaching strategy is significantly influenced by team composition and learner proficiency, highlighting the importance of real-time assessment of learner proficiency and adapting teaching approaches based on learner proficiency for optimal teaching outcomes.

Different people respond to feedback and guidance in different ways. Their preferences may even change depending on their mood, fatigue, physical health, etc. We present a robot exercise coach that provides both verbal and nonverbal feedback. We first introduce an exercise evaluation method where the camera feed from the robot is used to evaluate how well people perform exercises. We then present a multi-modal feedback controller that uses the exercise evaluation to respond with verbal and nonverbal feedback in different styles (firm and encouraging). Our user study found that participants have significantly different performances and subjective experiences with the different styles. We also found differences in how participants with different preferences for the styles perform with the different styles. These results show that varying feedback styles has an impact and builds the basis for a robot that adapts its style in real-time to personalize to the individual.

A majority of the courses on autonomous systems focus on robotics, despite the growing use of autonomous agents in a wide spectrum of applications, from smart homes to intelligent traffic control. Our goal in designing a new senior-level undergraduate course is to teach the integration of a variety of AI techniques in uncertain environments, without the dependence on topics such as robotic control and localization. We chose the application of an autonomous greenhouse to frame our discussions and our student projects because of the greenhouse's self-contained nature and objective metrics for successfully growing plants. We detail our curriculum design, including lecture topics and assignments, and our iterative process for updating the course over the last four years. Finally, we present some student feedback about the course and opportunities for future improvement.

This workshop aims to bring together researchers to explore and identify ways in which human-robot collaboration can reap the benefits of modern machine learning. The intended outcome is a roadmap that identifies key milestones that will lead us towards fluent effective human-robot teaming. In addition to focus groups and creative brainstorming exercises, this workshop will comprise invited talks, contributed paper talks, a poster session, and a debate. The papers, talks, posters, and roadmap will be made publicly available on our website: https://sites.google.com/view/mlhrc-hri-2022/home.

To collaborate well with robots, we must be able to understand their decision making. Humans naturally infer other agents' beliefs and desires by reasoning about their observable behavior in a way that resembles inverse reinforcement learning (IRL). Thus, robots can convey their beliefs and desires by providing demonstrations that are informative for a human learner's IRL. An informative demonstration is one that differs strongly from the learner's expectations of what the robot will do given their current understanding of the robot's decision making. However, standard IRL does not model the learner's existing expectations, and thus cannot do this counterfactual reasoning. We propose to incorporate the learner's current understanding of the robot's decision making into our model of human IRL, so that a robot can select demonstrations that maximize the human's understanding. We also propose a novel measure for estimating the difficulty for a human to predict instances of a robot's behavior in unseen environments. A user study finds that our test difficulty measure correlates well with human performance and confidence. Interestingly, considering human beliefs and counterfactuals when selecting demonstrations decreases human performance on easy tests, but increases performance on difficult tests, providing insight on how to best utilize such models.

Human and robot partners increasingly need to work together to perform tasks as a team. Robots designed for such collaboration must reason about how their task-completion strategies interplay with the behavior and skills of their human team members as they coordinate on achieving joint goals. Our goal in this work is to develop a computational framework for robot adaptation to human partners in human-robot team collaborations. We first present an algorithm for autonomously recognizing available task-completion strategies by observing human-human teams performing a collaborative task. By transforming team actions into low dimensional representations using hidden Markov models, we can identify strategies without prior knowledge. Robot policies are learned on each of the identified strategies to construct a Mixture-of-Experts model that adapts to the task strategies of unseen human partners. We evaluate our model on a collaborative cooking task using an Overcooked simulator. Results of an online user study with 125 participants demonstrate that our framework improves the task performance and collaborative fluency of human-agent teams, as compared to state of the art reinforcement learning methods.

Nonverbal communication in the field of education can allow teachers to emotionally support their students and improve educational experience and performance. Robot nonverbal movements have been shown to improve both subjective experiences and task performance, and this work investigates whether affective robot behavior can improve human learning. This is tested using an online sorting game where players learn easy or difficult rules, aided by robot feedback videos that contain either neutral or affective movements. Results indicate that affective robot behavior improves learning of the sorting rules and reduces the perceived difficulty of the task. Extensions include expanding the features used to determine the robot feedback and increasing the possible robot motions to create a rich set of robot feedback options to personalize the education experience further for the student.

The success of data scientists in developing machine learning models is contingent on an iterative development process for detecting patterns in data, finding and extracting useful features, and maximizing their model’s performance. However, it is often the case that they struggle during model development and become stuck and unable to make significant progress. We collected qualitative and quantitative data from the workflow of data scientists that allow us to learn from and examine such moments of stuckness. We used this data to develop a model for predicting stuckness based on real-time indicators, such as code artifacts, and then used the model to develop an innovative algorithm that determines precisely when a potential stuckness intervention should occur: as close as possible to the beginning of actual stuckness. Our algorithm’s performance indicates the potential efficacy of predicting data scientist stuckness algorithmically under real-world circumstances and for real-world needs.

As development of robots with the ability to self-assess their proficiency for accomplishing tasks continues to grow, metrics are needed to evaluate the characteristics and performance of these robot systems and their interactions with humans. This proficiency-based human-robot interaction (HRI) use case can occur before, during, or after the performance of a task. This article presents a set of metrics for this use case, driven by a four-stage cyclical interaction flow: (1) robot self-assessment of proficiency (RSA), (2) robot communication of proficiency to the human (RCP), (3) human understanding of proficiency (HUP), and (4) robot perception of the human’s intentions, values, and assessments (RPH). This effort leverages work from related fields including explainability, transparency, and introspection, by repurposing metrics under the context of proficiency self-assessment. Considerations for temporal level (a priori, in situ, and post hoc) on the metrics are reviewed, as are the connections between metrics within or across stages in the proficiency-based interaction flow. This article provides a common framework and language for metrics to enhance the development and measurement of HRI in the field of proficiency self-assessment.

Human-in-the-loop Machine Learning (HIL-ML) is a widely adopted paradigm for instilling human knowledge in autonomous agents. Many design choices influence the efficiency and effectiveness of such interactive learning processes, particularly the interaction type through which the human teacher may provide feedback. While different interaction types (demonstrations, preferences, etc.) have been proposed and evaluated in the HIL-ML literature, there has been little discussion of how these compare or how they should be selected to best address a particular learning problem. In this survey, we propose an organizing principle for HIL-ML that provides a way to analyze the effects of interaction types on human performance and training data. We also identify open problems in understanding the effects of interaction types.

The ability to learn from large quantities of complex data has led to the development of intelligent agents such as self-driving cars and assistive devices. This data often comes from people via interactions such as labeling, providing rewards and punishments, and giving demonstrations or critiques. However, people's ability to provide high-quality data can be affected by human factors of an interaction, such as induced cognitive load and perceived usability. We show that these human factors differ significantly between interaction types. We first formalize interactions as a Markov Decision Process, and construct a taxonomy of these interactions to identify four archetypes: Showing, Categorizing, Sorting, and Evaluating. We then run a user study across two task domains. Our findings show that Evaluating interactions are more cognitively loading and less usable than the others, and Categorizing and Showing interactions are the least cognitively loading and most usable.

End-users' trust in automated agents is important as automated decision-making and planning is increasingly used in many aspects of people's lives. In real-world applications of planning, multiple optimization objectives are often involved. Thus, planning agents' decisions can involve complex tradeoffs among competing objectives. It can be difficult for the end-users to understand why an agent decides on a particular planning solution on the basis of its objective values. As a result, the users may not know whether the agent is making the right decisions, and may lack trust in it. In this work, we contribute an approach, based on contrastive explanation, that enables a multi-objective MDP planning agent to explain its decisions in a way that communicates its tradeoff rationale in terms of the domain-level concepts. We conduct a human subjects experiment to evaluate the effectiveness of our explanation approach in a mobile robot navigation domain. The results show that our approach significantly improves the users' understanding, and confidence in their understanding, of the tradeoff rationale of the planning agent.

Performing detailed work on objects requires precise localization. Currently humans aid machines in localization either by direct operation, or implicitly by designing a sequence of actions a robot follows. Our approach to automate localization is to reason over many potential actions, perform the best information gathering action, and then use the measurement obtained to update a non-Gaussian belief. We propose a method for autonomous localization of objects with initial 6DOF uncertainty capable of reasoning about and performing measurements with low uncertainty and arbitrary error models. Surprisingly, common methods capable of modeling arbitrary belief distributions perform poorly as measurement uncertainty decreases, so we modify a particle filter to handle these accurate measurements produced by tactile or laser sensors. We then show how the expected information gain of the proposed measurement can be calculated efficiently from these particles. We present experiments, both in simulation and on hardware, that show our method is both fast and accurate.

In this paper, we propose a touch-based localization approach for a potentially large and complex object with multiple internal degrees of freedom. Should a task only require a partial localization of the object, our method selects the appropriate information gathering actions to register the desired features. We use probabilistic methods to reason over the distribution of the estimated object poses in the 6-DOF configuration space. We introduce the datum-based particle filter to handle intrinsic tolerances between each of the sections of the object. We describe two alternative methods for the particle filter system: one using the full joint belief and the other reasonably simplifying the belief to achieve a better ability to scale. We present simulation results for both proposed methods to show the advantages of our approaches.

Dancers are human Expressive Motion experts and could theoretically help robots communicate their state to people, e.g., rushed, confused, curious. The problem is twofold: first, dancers are trained in human-motion whereas many robots are non-anthropomorphic, and second, most dancers are not programmers. This is where the present interface is useful: the robot demos a batch of motions, in person, and the dancer, who knows expressive motion when she sees it, rates each path's success at communicating a particular state. Using an evolutionary algorithm, the interface - where feedback is recorded on the robot's screen and motion is demonstrated via the robot - calculates a new batch of motions that explore variations of the top-rated paths from the previous generation. This approach addresses the challenges of visualizing the expressive potential of non-anthropomorphic robots, while also ensuring path characteristics are reproducible via the robot's motion controller. The purpose of the interface is to help a non-expert negotiate a high-dimensional space of robot motion expression. Thus, it also has interactive functionality enabling users to freeze a feature value they like, or reset all features to begin again. To illustrate the system, this paper includes the results of two dancers designing motions for an omni-directional mobile robot, showing convergence with every generation. In reality, motion designers may have many authoring styles - exploring multiple solutions before honing in, or being satisfied easily versus getting each detail exactly right. By combining human-in-the-loop machine learning with direct authoring, we create a kinetic conversation between the robot and the dancer, and gain the ability to model knowledge from complementary fields.

Expressive motion refers to movements that help convey an agent's attitude towards its task or environment. People frequently use expressive motion to indicate internal states such as emotion, confidence, and engagement. Robots can also exhibit expressive motion, and studies have shown that people can legibly interpret such expressive motion. Mimicry involves imitating the behaviors of others, and has been shown to increase rapport between people. The research question addressed in this study is how robots mimicking the expressive motion of children affects their interaction with dancing robots. The paper presents our approach to generating and characterizing expressive motion, based on the Laban Efforts System and the results of the study, which provides both significant and suggestive evidence to support that such mimicry has positive effects on the children's behaviors.

Sudden weight gain in patients living with Congestive Heart Failure (CHF) is often an indication that the individual is retaining fluid, which often means that patient's heart has weakened leading to increased risk of kidney or cardiac failure. Clinical interventions can be made at this stage, leading to better outcomes, however it is essential that the interventions take place before the patient's health declines too drastically. In this work, we present a latent variable autoregression model that tracks patient weight and blood pressure over time, allowing us to predict weight values into the future. We are also able to model continuous heart-rate signals and evaluate a subject's response to physical activity. This allows us to detect signs of health decline days earlier than existing rule-based systems, leading to the possibility of earlier clinical interventions, potentially preventing deadly medical emergencies.

Functional robots are an increasing presence in shared human-machine environments. Humans efficiently parse motion expressions, gaining an immediate impression of an agent's current action and state. Past work has shown that motion can effectively reveal a robot's current task objective to bystanders and collaborators, however, the layering of expression on pre-existing robot task motions has yet to be explored. Rather than showing us what the robot is doing, these layered motion characteristics leverage the how of the task motions to convey additional robot attitudes, e.g., confidence, adherence to deadline or flexibility of attention. To lay the foundations for this objective, we adapt the Laban Efforts, a system from dance and acting training in use for over 50 years. We operationalize features representing the four Laban Efforts (Time, Space, Weight, and Flow) to the movements of a 2-DOF Nao head and a 4-DOF Keepon robot during simple dance and look-for-someone behaviors. Using online survey, we collect 1028 motion ratings for 72 robot motion videos depicting contrasting Effort motion examples. We achieve statistically significant legibility results for all four Effort implementations. Even without human degrees of freedom, we find that robot motion patterns can convey complex expressions to people.

Expressive motion can situate a robot's attitude in its task motions, illustrating real-time reactions. Inspired by acting movement training, we construct path shape features that layer expression into a mobile robot's motion traversal. Our video-study results show that simple variations of path shape and orientation can influence human perceptions of a robot's task, focus, and confidence. We further find that sequencing path features is a useful way to create expressions that are pinpointed in time without requiring changes in velocity. Our quantitative features represent the Laban Space Effort: using path shape and orientation along the path to communicate the direct or indirect attitude of the robot toward its target destination (acting vocabulary italicized). These features illustrate expressive or stylistic aspects of the robot's inner state, filling a gap in the pre-existing literature that has mostly focused on task legibility. Our future work will evaluate temporal and spatial robot motion features in explicit interaction contexts.

In this paper we present a new method for understanding discussions between students in MOOC forums. In particular, we introduce a machine learning method for discovering instances in which a response relation exists between a pair of posts in a forum thread, for example when one student provides the answer to a question or comments on something another student previously said. Research has shown that understanding conversational structure between students is paramount to evaluating the productivity of the collaboration and estimating outcomes. However, previous methods often rely on human supplied dialogue act labels or discourse parsing algorithms requiring large labeled datasets. Our method, which utilizes a fast, exact optimization process known as spectral optimization, does not require manually annotated training data and is highly scalable and generalizable. Empirical results are given using real world datasets consisting of conversations between students participating in Coursera courses, and we see predictive accuracy above 90% - nearing the human inter-annotator agreement rate for these datasets.

We present a method to learn context-dependent outcomes of behaviors in unstructured indoor environments. The idea is that certain features in the environment may be predictive of differences in outcomes, such as how long a mobile robot takes to traverse a corridor. Doing so enables the robot to plan more effectively, and also be able to interact with people more effectively by more accurately predicting when its plans may take longer to execute or may be likely to fail. We use a node-and-edge based map of the environment and treat the traversal time of the robot for each edge as a random variable to be characterized. The first step is to determine whether the distribution of the random variable is multimodal and, if so, we learn to classify the modes using a hierarchy of plan-time features (e.g., time of the day, day of the week) and run-time features (observations of recent traversal times through other corridors). We utilize a cascading regression system that first estimates which mode of the traversal distribution we expect the robot to observe, and then predict the actual traversal time through a corridor. On average, our method produces a mean residual error of less than 2.7 seconds.

Assembly of large structures requires large fixtures, often referred to as monuments. Their cost and massive size limit flexibility and scalability of the manufacturing process. Numerous small mobile robots can replace these large structures and, therefore, replicate the efficiency of the assembly line with far more flexibility. An assembly line made up of mobile manipulators can easily and rapidly be reconfigured to support scalability and a varied product mix, while allowing for near optimal resource assignment. The challenge to using small robots in place of monuments is making their joint behavior precise enough to accomplish the task and efficient enough to execute subtasks in a reasonable period of time. In this paper, we describe a set of techniques that we combine to achieve the necessary precision and overall efficiency to build a large structure. We describe and demonstrate these techniques in the context of a testbed we implemented for assembling a wing ladder.

Modeling the effects of actions based on the state of the world enables robots to make intelligent decisions in different situations. However, it is often infeasible to have globally accurate models. Task performance is often hindered by discrepancies between models and the real world, since the true outcome of executing a plan may be significantly worse than the expected outcome used during planning. Furthermore, expectations about the world are often stochastic in robotics, making the discovery of model-world discrepancies non-trivial. We present an execution monitoring framework capable of finding statistically significant discrepancies, determining the situations in which they occur, and making simple corrections to the world model to improve performance. In our approach, plans are initially based on a model of the world that is only as faithful as computational and algorithmic limitations allow. Through experience, the monitor discovers previously unmodeled modes of the world, defined as regions of a feature space in which the experienced outcome of a plan deviates significantly from the predicted outcome. The monitor may then make suggestions to change the model to match the real world more accurately. We demonstrate this approach on the adversarial domain of robot soccer: we monitor pass interception performance of potentially unknown opponents to try to find unforeseen modes of behavior that affect their interception performance.

In our experiment, two autonomously moving costumed robots visit 256 offices during a `reverse' trick-or-treating task close to Halloween. Our behavioral data supports the idea that people interpret a robot's non-verbal cues, as the robots' costuming and baskets of candy seem to have communicated an implicit offer of candy. In fact, one third of our detection instances occurred during robot transit, i.e., while the robots were making no verbal offer. We find that candy accessibility dominates any social influence of robot orientation and that robot speed influences both whether people will interrupt a robot in transit (slow more interruptible) and whether they will respond to its verbal offer (fast more salient).

Robots are possible candidates for performing tasks as helpers in activities of daily living in the future: working as a receptionist is one possible employment. However, the way the receptionist robot should appear, sound and behave needs to be investigated carefully, in order to design a robot which is accepted and perceived in a positive way by common users. This paper describes a study on anthropomorphism of a receptionist robot made for Brazilian people depending on the appearance and on the voice of the receptionist. This experiment was preceded by a preliminary survey about expectation of people regarding receptionists. The main experiment consisted in having Brazilian people interacting with a conversational agent and with a humanoid robot through a video conference. The two receptionists are not only different in physical appearance, but in the sound of the voice, too, which can be either human-like or robotic sound. The two receptionists gave indications to the participants to reach rooms where they could evaluate the receptionists through questionnaires concerning anthropomorphism and uncanniness among other concepts. The results gathered from both experiments provide useful hints to design a receptionist robot.

This paper presents an online algorithm for early detection of anomalies in robot execution, where the anomalies occur in a particular region of the robot's state space. Assuming that a model of normal execution is given, the algorithm detects regions of space where data significantly deviate from normal. It achieves this by focusing optimization over a fixed-parameter family of shapes to find the one among them that is most likely anomalous, and then using this region to decide whether execution is anomalous. Experiments using synthetic and real robot data support the effectiveness of the approach.

For service robots operating in indoor environments, the crucial task of navigation is often complicated by the presence of people. Simply treating humans in the environment as additional (often moving) obstacles can violate the complex set of social rules by which people navigate around each other. In contrast, emulating human behavior and navigating in a socially appropriate manner could positively affect people's comfort with a robot's presence and motion. We present a method of generating social paths for a robot to approach a person based on a small amount of human data. We also conducted a study in which a robot approached participants using both these social paths and straight-line, nonsocial paths. We found that both approaches were rated comparably when the robot approached from the participant's front or side, but the social approach was significantly preferred when the robot came from behind the participant.

There is a saying that 95% of communication is body language, but few robot systems today make effective use of that ubiquitous channel. Motion is an essential area of social communication that will enable robots and people to collaborate naturally, develop rapport, and seamlessly share environments. The proposed work presents a principled set of motion features based on the Laban Effort system, a widespread and extensively tested acting ontology for the dynamics of “how” we enact motion. The features allow us to analyze and, in future work, generate expressive motion using position (x, y) and orientation (theta). We formulate representative features for each Effort and parameterize them on expressive motion sample trajectories collected from experts in robotics and theater. We then produce classifiers for different “manners” of moving and assess the quality of results by comparing them to the humans labeling the same set of paths on Amazon Mechanical Turk. Results indicate that the machine analysis (41.7% match between intended and classified manner) achieves similar accuracy overall compared to a human benchmark (41.2% match). We conclude that these motion features perform well for analyzing expression in low degree of freedom systems and could be used to help design more effectively expressive mobile robots.

In this paper we analyze joint attention between a robot that presents features of its surroundings and its human audience. In a statistical analysis of hand-coded video data, we find that the robot's physical indications lead to a greater attentional coherence between robot and humans than do its verbal indications.We also find that aspects of how the tour group participants look at robot-indicated objects, including when they look and how long they look, can provide statistically significant correlations with their self-reported engagement scores of the presentations. Higher engagement would suggest a greater degree of interest in, and attention to, the material presented. These findings will seed future gaze tracking systems that will enable robots to estimate listeners' state. By tracking audience gaze, our goal is to enable robots to cater the type of content and manner of its presentation to the preferences or educational goals of a particular crowd, e.g. in a tour guide, classroom or entertainment setting.

As an early behavioral study of what non-verbal features a robot tourguide could use to analyze a crowd, personalize an interaction and/or maintain high levels of engagement, we analyze participant gaze statistics in response to a robot tour guide's deictic gestures. There were thirty-seven participants overall split into nine groups of three to five people each. In groups with the lowest engagement levels aggregate gaze responses in response to the robot deictic gesture involved the fewest total glance shifts, least time spent looking at indicated object and no intra-participant gaze. Our diverse participants had overlapping engagement ratings within their group, and we found that a robot that tracks group rather than individual analytics could capture less noisy and often stronger trends relating gaze features to self-reported engagement scores. Thus we have found indications that aggregate group analysis captures more salient and accurate assessments of overall humans-robot interactions, even with lower resolution features.

Identification of regions in space that a robotic manipulator can reach in a given amount of time is important for many applications, such as safety monitoring of industrial manipulators and trajectory and task planning. However, due to the high-dimensional configuration space of many robots, reasoning about possible physical motion is often intractable. In this paper, we propose a novel method for creating a reachability grid, a voxel-based representation that estimates the minimum time needed for a manipulator to reach any physical location within its workspace. We use up to second-degree constraints on joint motion to model motion limits for each joint independently, followed by successive voxel approximations to map these limits on to the robot's physical workspace. Results using a simulated manipulator indicate that our method can produce accurate reachability grids in real-time, even for robots with many degrees of freedom. Furthermore, errors are almost exclusively biased towards producing more optimistic reachability estimates, which is a desirable characteristic for many applications.

As robots are utilized in a growing number of applications, the ability to teach them to perform tasks safely and accurately becomes ever more critical. Programming by demonstration offers an expressive means for teaching while being accessible to domain experts who may be novices in robotics. This work investigates a programming by demon- stration approach to learning motion trajectories for robotic manipulator tasks. Using a graph constructed to determine correspondences between multiple imperfect demonstrations, the robot learner plans novel trajectories that safely and smoothly generalize the teacher's behavior, while attenuating those imperfections. The learner also actively detects instances of diverging strategy between examples, requesting advice for resolving these ambiguities. We demonstrate our approach in example domains with a 7 degree-of-freedom manipulator.

As mobile robots become better equipped to autonomously navigate in human-populated environments, they need to become able to recognize internal and external factors that may interfere with successful motion execution. Even when these robots are equipped with appropriate obstacle avoidance algorithms, collisions and other forms of motion interference might be inevitable: there may be obstacles in the environment that are invisible to the robot's sensors, or there may be people who could interfere with the robot's motion. We present a Hidden Markov Model-based model for detecting such events in mobile robots that do not include special sensors for specific motion interference. We identify the robot observable sensory data and model the states of the robot. Our algorithm is motivated and implemented on an omnidirectional mobile service robot equipped with a depth-camera. Our experiments show that our algorithm can detect over 90% of motion interference events while avoiding false positive detections.

Current manufacturing practices require complete physical separation between people and active industrial robots. These precautions ensure safety, but are inefficient in terms of time and resources, and place limits on the types of tasks that can be performed. In this paper, we present a real-time, sensor-based approach for ensuring the safety of people in close proximity to robots in an industrial workcell. Our approach fuses data from multiple 3D imaging sensors of different modalities into a volumetric evidence grid and segments the volume into regions corresponding to background, robots, and people. Surrounding each robot is a danger zone that dynamically updates according to the robot's position and trajectory. Similarly, surrounding each person is a dynamically updated safety zone. A collision between danger and safety zones indicates an impending actual collision, and the affected robot is stopped until the problem is resolved. We demonstrate and experimentally evaluate the concept in a prototype industrial workcell augmented with stereo and range cameras.

Evidence grids are a popular representation for fused data from multiple sensors. Previous attempts at back ground subtraction within evidence grids either do so prior to sensor fusion or do so naively, simply ignoring any cells with a high background occupancy probability. A key weakness of these approaches is that they cannot reason about interiors of objects or other unobserved regions. Recognizing and removing solid object interiors is important for any application that must be able to differentiate between occupied and unknown space after background subtraction. In this paper, we propose accessibility analysis as a method for the removal of interior regions. We then present and compare two approaches for performing background subtraction with accessibility analysis in evidence grids. Performance is measured using a 3D evidence grid in a test bed for a sensing system designed for use in safety monitoring of an automated assembly workcell. Within the parameters of the present study, both techniques allow for precise detection of foreground objects while fully removing background objects. Subtraction runs in near real-time, even for large grids.

In this paper, a modelling approach is described that represents human-robot social interactions as partially observable Markov decision processes (POMDPs). In these POMDPs, the intention of the human is represented as an unobservable part of the state space, and the robot's own intentions are expressed through the rewards. The state transition structure for the models is created using action rules that capture the effects of the robot's actions, relate the human's behavior to their intentions, and describe the changing state of the environment. State transitions are modified using data from humans interacting with other humans. The policies obtained by solving these models are used to control a robot in a socially situated task with a human partner. These interactions are compared to those of human pairs performing the same task, demonstrating that this approach produces policies that exhibit natural and socially appropriate behavior.

Hala is a bilingual (Arabic and English) culturally-sensitive robot receptionist located at Carnegie Mellon University in Qatar. We report results from Hala's deployment by comparing her English dialogue corpus to that of a similar monolingual robot (named "Tank") located at CMU's Pittsburgh campus. Specifically, we compare the average number of turns per interaction, duration of interactions, frequency of interactions with personal questions, rate of non-understandings, and rate of thanks after the robot's answer. We provide possible explanations for observed similarities and differences and highlight potential cultural implications on the interactions.

Programming by demonstration is an attractive model for allowing both experts and non-experts to command robots' actions. In this work, we contribute an approach for learning precise reaching trajectories for robotic manipulators. We use dimensionality reduction to smooth the example trajectories and transform their representation to a space more amenable to planning. Key to this approach is the careful selection of neighboring points within and between trajectories. This algorithm is capable of creating efficient, collision-free plans even under typical real-world training conditions such as incomplete sensor coverage and lack of an environment model, without imposing additional requirements upon the user such as constraining the types of example trajectories provided. Experimental results are presented to validate this approach.

Similarly, User models can be useful for improving dialogue management. In this paper we analyze human-robot dialogues that occur during uncontrolled interactions and estimate relations between the initial dialogue turns and patterns of discourse that are indicative of such user traits as persistence and politeness. The significant effects shown in this preliminary study suggest that initial dialogue turns may be useful in modeling a user's interaction style.

Traditional industrial robots have been widely used in automotive manufacturing for nearly 30 years. However, there have been very few attempts to automate mobile robotic systems for final assembly operations, despite their potential for high flexibility and capability. This paper focuses on methods of tracking a dynamic moving vehicle that is similar to the vehicle body on a moving assembly line. We have investigated two tracking methods, one using a laser scanner and the other using a visual fiducial marker. We have also studied the tracking performance of a mobile base using the pure pursuit algorithm with low pass filtering. Experimental results are presented to illustrate the remaining main challenges in achieving robotic assembly on moving assembly lines.

This paper presents a method for improving the runtime of an optimal heuristic path planner (A*) so that it can run repeatedly, in real-time, in a dynamic environment. This is necessary for mobile robots navigating in dynamic environments that have moving obstacles with associated costs, such as personal space around people or buffer zones around dangerous vehicles. Our approach is to modify the search space used by the A* algorithm, increasing the size of grid cells further from the robot. This approach relies on the notion that only the area closest to the robot needs to be searched carefully; areas further from the robot can be searched more coarsely. Because the planner is assumed to run repeatedly as the robot moves, the robot will always have a fine-grained path defined for its next action. We have experimentally verified in simulation that this algorithm can be run in real-time and produces paths that are comparable to full-resolution planning.

This paper introduces the COMPANION framework: a constraint-optimizing method for person-acceptable navigation. In this framework, human social conventions, such as personal space and tending to one side of hallways, are represented as constraints on the robot's navigation. These constraints are accounted for at the global planning level. In this paper, we present the rationale for, and implementation of, this framework, and we describe the experiments we have run in simulation to verify that the method produces human-like behavior in a mobile robot. Our approach is novel in that it can express an arbitrary number of social conventions and explicitly accounts for these conventions in the planning phase.

Dialogues with robots frequently exhibit social dialogue acts such as greeting, thanks, and goodbye. This opens the opportunity of using these dialogue acts for dialogue management, in particular for detecting misunderstandings. Our corpus analysis shows that the social dialogue acts have different scopes of their associations with the discourse features within the dialogue: greeting in the user's first turn is associated with such distant, or global, features as the likelihood of having questions answered, persistence, and ending with bye. The user's thanks turn, on the other hand, is strongly associated with the helpfulness of the preceding robot's answer. We therefore interpret the greeting as a component of a user model that can provide information about the user's traits and be associated with discourse features at various stages of the dialogue. We conduct a detailed analysis of the user's thanking behavior and demonstrate that user's thanks can be used in the detection of unhelpful robot's answers. Incorporating the greeting information further improves the detection. We discuss possible applications of this work for human-robot dialogue management.

Human social behavior is rhythmic, and synchrony plays an important role in coordinating and regulating our interactions. We are developing technology that allows the robot Keepon to perceive and behave rhythmically, and to synchronize its dancing behaviors to music or to children's movement as perceived using pressure sensors. We present two experiments in which Keepon dances with children to music, and in which the robot's rhythmic attention and role of leader or follower are manipulated in order to examine the effects on engagement and rhythmic synchrony. We found that children can assume the roles of leader or follower in a rhythmic interaction, that followers indeed tend to synchronize with the robot's movements, and that the role of follower causes the children to more closely follow a musical rhythm.

Proactive replanning attempts to predict scheduling problems or opportunities and adapt to them throughout a schedule's execution. By continuously predicting a task's remaining duration, a proactive replanner is able to accommodate upcoming problems or opportunities before they manifest themselves. We have developed a kernel density estimation-based method for predicting a task's duration distribution as it executes, and have integrated our prediction algorithm with an existing planner based on heuristic repair. Our predictor allows the planner to anticipate problems, or opportunities, early enough to avoid, or take advantage of, them, resulting in executed schedules that score significantly higher on a number of metrics. We have evaluated a limited form of our approach in simulation, and present the results of our experiments. The addition of duration prediction resulted in a 11.1% improvement in average reward. Compared with an omniscient planner, this is 45.0% of the maximum possible improvement.

The capability to assemble structures is fundamental to the use of robotics in precursor missions in orbit and on planetary surfaces. We have performed autonomous assembly in neutral buoyancy of elements of a space truss whose mating components require positioning tolerances of the same order of magnitude as the noise in the sensor systems used for the docking. Numerous trade-offs, design decisions, and innovations were made during the development of the assembly system in order to both reduce and compensate for the sensor noise. By using relative positioning, decoupling sensing and manipulation, caching high-quality position estimates, and developing a new waypoint-completion metric, we were able to reduce sensor noise to the sub-millimeter level and autonomously assemble components with millimeter tolerances. In this paper, we discuss our approaches to the problem and report the results of a series of autonomous assembly operations.

We are developing robots with socially appropriate spatial skills not only to travel around or near people, but also to accompany people side-by-side. As a step toward this goal, we are investigating the social perceptions of a robot's movement as it follows behind a person. This paper discusses our laser-based person-tracking method and two different approaches to person-following: direction-following and path-following. While both algorithms have similar characteristics in terms of tracking performance and following distances, participants in a pilot study rated the direction-following behavior as significantly more human-like and natural than the path-following behavior. We argue that the path-following method may still be more appropriate in some situations, and we propose that the ideal person-following behavior may be a hybrid approach, with the robot automatically selecting which method to use.

In many robotic domains, efficiency is an important component of task execution. One way to improve task efficiency is to lessen the overhead of beginning a task by making sure the necessary agents are near the task site when execution begins, minimizing travel time delays - in other words, pre-positioning agents for their future tasks. In static, certain domains, this can easily be done in advance and incorporated into the initial plan. In dynamic domains such as search and rescue, however, there is not enough certainty about task execution to plan for this ahead of time. To address this, we present here a planner that adds pre-positioning to a plan during execution. The planner strategically positions groups of idle robots whose future task assignments are uncertain in order to minimize travel time by the group as a whole once its members are allocated tasks. Because this planner must run in real time, we present five versions of the planning algorithm, addressing the trade-off of computation time and solution quality that results. We then show that by adding in this type of planning, the overhead of beginning a task can be reduced by up to 90%.

This paper describes a new extension to the rapidly-exploring random tree (RRT) path planning algorithm. The particle RRT algorithm explicitly considers uncertainty in its domain, similar to the operation of a particle filter. Each extension to the search tree is treated as a stochastic process and is simulated multiple times. The behavior of the robot can be characterized based on the specified uncertainty in the environment, and guarantees can be made as to the performance under this uncertainty. Extensions to the search tree, and therefore entire paths, may be chosen based on the expected probability of successful execution. The benefit of this algorithm is demonstrated in the simulation of a rover operating in rough terrain with unknown coefficients of friction

This paper reports on the results of a long-term experiment in which a social robot's facial expressions were changed to reflect different moods. While the facial changes in each condition were not extremely different, they still altered how people interacted with the robot. On days when many visitors were present, average interactions with the robot were longer when the robot displayed either a "happy" or a "sad" expression instead of a neutral face, but the opposite was true for low-visitor days. The implications of these findings for human-robot social interaction are discussed.

This paper presents a robot search task (social tag) that uses social interaction, in the form of asking for help, as an integral component of task completion. We define socially distributed perception as a robot's ability to augment its limited sensory capacities through social interaction.

The problems of human detection, tracking, and attention recognition can be solved more effectively by integrating multiple sensory modalities, such as vision and range data. We present a system that uses a laser range scanner and a single camera to detect and track people, and to classify their attention relative to a socially interactive robot.

We are interested in the problem of a human operator who has to respond to requests for help from an autonomous robotic construction team. A difficult aspect of this problem is gaining an awareness of the requesting robot’s situation quickly enough to avoid slowing the whole team down. One approach to speeding the initial acquisition of situational awareness is to maintain a buffer of data, and play it back for the human when their help is needed. The paper reports on an experiment to determine the proper composition and length of this buffer for our domain of multi-robot construction. The experiments show that 5- 10 seconds of one raw video feed in combination with a processed display led to the fastest operator attainment of situational awareness. We draw several conclusions from this experiment, which may generalize to other scenarios.

An important mode of human-robot interaction is teleoperation, in which a human operator directly controls a robot via hardware such as a joystick or mouse. Such control is not always easy, however, as the viewpoint of the human, the alignment of the input device, and the local coordinate frame of the robot are rarely all aligned. These discrepancies force the user to reconcile the involved coordinate frames during teleoperation. Therefore, the choice of coordinate frames is critical since an unintuitive coordinate frame mapping will likely lead to higher mental workload and reduced efficiency. We discuss this concern, describe the various difficulties involved with natural remote teleoperation of a robot, and report experiments that demonstrate the effects of using different frames of reference on task performance and user mental workload

Humans use expressions of emotion in a very social manner, to convey messages such as "I'm happy to see you" or "I want to be comforted," and people's long-term relationships depend heavily on shared emotional experiences. We believe that for robots to interact naturally with humans in social situations they should also be able to express emotions in both short-term and long-term relationships. To this end, we have developed an affective model for social robots. This generative model attempts to create natural, human-like affect and includes distinctions between immediate emotional responses, the overall mood of the robot, and long-term attitudes toward each visitor to the robot. This paper presents the general affect model as well as particular details of our implementation of the model on one robot, the Roboceptionist

The Peer-To-Peer Human-Robot Interaction (P2P-HRI) project is developing techniques to improve task coordination and collaboration between human and robot partners. Our work is motivated by the need to develop effective human-robot teams for space mission operations. A central element of our approach is creating dialogue and interaction tools that enable humans and robots to flexibly support one another. In order to understand how this approach can influence task performance, we recently conducted a series of tests simulating a lunar construction task with a human-robot team. In this paper, we describe the tests performed, discuss our initial results, and analyze the effect of intervention on task performance.

Recent research in human-robot interaction has investigated the concept of Sliding, or Adjustable, Autonomy, a mode of operation bridging the gap between explicit teleoperation and complete robot autonomy. This work has largely been in single-agent domains-involving only one human and one robot-and has not examined the issues that arise in multiagent domains. We discuss the issues involved in adapting Sliding Autonomy concepts to coordinated multiagent teams. In our approach, remote human operators have the ability to join, or leave, the team at will to assist the autonomous agents with their tasks (or aspects of their tasks) while not disrupting the team's coordination. Agents model their own and the human operator's performance on subtasks to enable them to determine when to request help from the operator. To validate our approach, we present the results of two experiments. The first evaluates the human/multirobot team's performance under four different collaboration strategies including complete teleoperation, pure autonomy, and two distinct versions of Sliding Autonomy. The second experiment compares a variety of user interface configurations to investigate how quickly a human operator can attain situational awareness when asked to help. The results of these studies support our belief that by incorporating a remote human operator into multiagent teams, the team as a whole becomes more robust and efficient

Direct human control of multi-robot systems is limited by the cognitive ability of humans to coordinate numerous interacting components. In remote environments, such as those encountered during planetary or ocean exploration, a further limit is imposed by communication bandwidth and delay. Market based planning can give humans a higher-level interface to multi-robot systems in these scenarios. Operators provide high level tasks and attach a reward to the achievement of each task. The robots then trade these tasks through a market based mechanism. The challenge for the system designer is to create bidding algorithms for the robots that yield high overall system performance. Opportunity cost provides a nice basis for such bidding algorithms since it encapsulates all the costs and benefits we are interested in. Unfortunately, computing it can be difficult. We propose a method of learning opportunity costs in market based planners. We provide analytic results in simplified scenarios and empirical results on our FIRE simulator, which focuses on exploration of Mars by multiple, heterogeneous rovers.

Valerie the roboceptionist is the most recent addition to Carnegie Mellon's social robots project. A permanent installation in the entranceway to Newell-Simon hall, the robot combines useful functionality - giving directions, looking up weather forecasts, etc. - with an interesting and compelling character. We are using Valerie to investigate human-robot social interaction, especially long-term human-robot "relationships". Over a nine-month period, we have found that many visitors continue to interact with the robot on a daily basis, but that few of the individual interactions last for more than 30 seconds. Our analysis of the data has indicated several design decisions that should facilitate more natural human-robot interactions.

We are developing a coordinated team of robots to assemble structures, a task that cannot be performed by any single robot. Even simple operations in this domain require complex interaction between multiple robots and the number of contingencies that must be addressed if the team is to act completely autonomously is prohibitively large. This scenario forces incorporation of a human operator. Ideally we would like a seamless interface between the robots and the operator such that the operator can interact with the system by helping it be more efficient or get out of a stuck condition or performing a task that the robots are not capable of themselves. We use an architecture that implements "sliding autonomy" to accomplish these goals. The system of robots can be fully autonomous as long as all is well. The system is capable of accepting input from the operator at any time, especially when it is unable to recover from a failure. We motivate this scenario with results from an extended series of experiments we have conducted with three robots that work together to dock both ends of a suspended beam. We show the difference in performance between a completely teleoperated system, a fully autonomous system, and one in which sliding autonomy has been incorporated.

We present an algorithm designed to solve the problem of maintaining communications within a group of robotic explorers. The rovers we consider are equipped with communication hardware that is effective only over a limited range and requires direct line of sight to function. The paper presents the algorithm used to solve this problem and some details of our implementation. We also present the results of an experimental analysis of the algorithm's performance characteristics in a simulated multi-rover environment.

We present an overview of the Coupled Layered Architecture for Robotic Autonomy. CLARAty develops a framework for generic and reusable robotic components that can be adapted to a number of heterogeneous robot platforms. It also provides a framework that will simplify the integration of new technologies and enable the comparison of various elements. CLARAty consists of two distinct layers: a functional layer and a decision layer. The functional layer defines the various abstractions of the system and adapts the abstract components to real or simulated devices. It provides a framework and the algorithms for low- and mid-level autonomy. The decision layer provides the system's high-level autonomy, which reasons about global resources and mission constraints. The decision layer accesses information from the functional layer at multiple levels of granularity. We also present some of the challenges in developing interoperable software for various rover platforms.

This paper presents several modifications to the basic rapidly-exploring random tree (RRT) search algorithm. The fundamental idea is to utilize a heuristic quality function to guide the search. Results from a relevant simulation experiment illustrate the benefit and drawbacks of the developed algorithms. The paper concludes with several promising directions for future research.

VHPOP is a partial order causal link (POCL) planner loosely based on UCPOP. It draws from the experience gained in the early to mid 1990's on flaw selection strategies for POCL planning, and combines this with more recent developments in the field of domain independent planning such as distance based heuristics and reachability analysis. We present an adaptation of the additive heuristic for plan space planning, and modify it to account for possible reuse of existing actions in a plan. We also propose a large set of novel flaw selection strategies, and show how these can help us solve more problems than previously possible by POCL planners. VHPOP also supports planning with durative actions by incorporating standard techniques for temporal constraint reasoning. We demonstrate that the same heuristic techniques used to boost the performance of classical POCL planning can be effective in domains with durative actions as well. The result is a versatile heuristic POCL planner competitive with established CSP-based and heuristic state space planners.

This paper presents the results of an experiment in human-robot social interaction. Its purpose was to measure the impact of certain features and behaviors on people's willingness to engage in a short interaction with a robot. The behaviors tested were the ability to convey expression with a humanoid face and the ability to indicate attention by turning towards the person that the robot is addressing. We hypothesized that these features were minimal requirements for effective social interaction between a human and a robot. We will discuss the results of the experiment and their implications for the design of socially interactive robots.

Describes a set of tools that allows a developer to instrument an autonomous control system to log data at run-time and then analyze that data to verify correct program behavior. Analysis is done using an interval logic that allows system engineers to express complex, temporal specifications to be checked against the logged data of the autonomous control program. A feature of both the logging and analysis is that they can work with distributed programs. All data is synchronized into a common database. The data logging tools and the interval logic are fully implemented. Results are given from a NASA distributed autonomous control system application.

This paper describes the navigation system used on a prototype sun-synchronous robot. Sun-synchrony is a concept that will enable exploration missions by solar-powered rovers that could last months or years. This paper presents the navigation algorithms developed for traversing natural terrain robustly. The novel elements of this work are the refinements necessary to transform laboratory-demonstrated technologies into a form useful for robust, Sun-synchronous exploration. Results of afield experiment in the Canadian Arctic, where the robot traversed 6.1km, 90% autonomously, are also presented.

Enables robot explorers to maximize the total information gained while minimizing costs such as driving, sensing and planning. The paper presents a general methodology for solving complex exploration tasks which employs multiple sources of information. The paper also develops a specific instantiation of the method to solve the exploration problem of creating a complete traversability map of an known region. Simulation results showing the solution of this exploration task are included.

The Science Autonomy System (SAS) is a hierarchical control architecture for exploration and in situ science that integrates sensing, navigation, classification and mission planning. The Nomad robot demonstrated the capabilities of the SAS during a January 2000 expedition to Elephant Moraine, Antarctica where it accomplished the first meteorite discoveries made by a robot. In the paper, the structure and functionality of the three-tiered SAS are detailed. Results and lessons learned are presented with a focus on important future research.

In this paper we consider the problem of exploring an unknown environment by a team of robots. As in single-robot exploration the goal is to minimize the overall exploration time. The key problem to be solved therefore is to choose appropriate target points for the individual robots so that they simultaneously explore different regions of their environment. We present a probabilistic approach for the coordination of multiple robots which, in contrast to previous approaches, simultaneously takes into account the costs of reaching a target point and the utility of target points. The utility of target points is given by the size of the unexplored area that a robot can cover with its sensors upon reaching a target position. Whenever a target point is assigned to a specific robot, the utility of the unexplored area visible from this target position is reduced for the other robots. This way, a team of multiple robots assigns different target points to the individual robots. The technique has been implemented and tested extensively in real-world experiments and simulation runs. The results given in this paper demonstrate that our coordination technique significantly reduces the exploration time compared to previous approaches.

Architectures form the backbone of complete robotic systems. The right choice of architecture can go a long way in facilitating the specification, implementation and validation of robotic systems. Conversely, of course, the wrong choice can make one's life miserable. We present some of the needs of robotic systems, describe some general classes of robot architectures, and discuss how different architectural styles can help in addressing those needs. The paper, like the field itself, is somewhat preliminary, yet it is hoped that it will provide guidance for those who use, or develop, robot architectures.

Autonomous planetary rovers operating in vast unknown environments must operate efficiently because of size, power and computing limitations. Recently, we have developed a rover capable of efficient obstacle avoidance and path planning. The rover uses binocular stereo vision to sense potentially cluttered outdoor environments. Navigation is performed by a combination of several modules that each "vote" for the next best action for the robot to execute. The key distinction of our system is that it produces globally intelligent behavior with a small computational resource - all processing and decision making are done on a single processor. These algorithms have been tested on our outdoor prototype rover, Bullwinkle, and have recently driven the rover 100 m at a speed of 15 cm/sec. In this paper we report on the extension on the systems that we have previously developed that were necessary to achieve autonomous navigation in this domain.

This paper presents experimental results of preliminary research into multi-robot coordination for construction tasks. Experiments demonstrate that an autonomous "roving eye" robot can provide feedback to a manipulator to align targets from a wider variety of situations than is possible with fixed cameras, without sacrificing the accuracy provided by cameras at close range. The roving eye changes its location autonomously based on current images of the manipulated object and target, always striving for the best view of the task.

Recent research results on mobile robot navigation systems make it promising to utilize them in service fields. But in order to utilize the robot in a peopled environment, it should recognize and respond to people's social behaviors. In this paper, we describe a social robot that stands in line as people do. Our system uses the concept of personal space for modeling a line of people and we have experimentally measured the actual size of the personal space when people form lines. The system employs stereo vision to recognize lines of people. We demonstrate our ideas with a mobile robot navigation system that can purchase a cup of coffee, even if people are waiting in line for service.

To be truly useful, mobile robots need to be fairly autonomous and easy to control. This is especially true in situations where multiple robots are used, due to the increase in sensory information and the fact that the robots can interfere with one another. The paper describes a system that integrates autonomous navigation, a task executive, task planning, and an intuitive graphical user interface to control multiple, heterogeneous robots. We have demonstrated a prototype system that plans and coordinates the deployment of teams of robots. Testing has shown the effectiveness and robustness of the system, and of the coordination strategies in particular.

While autonomous systems offer great promise in terms of capability and flexibility, their reliability is particularly hard to assess. This paper describes research to apply formal verification methods to languages used to develop autonomy software. In particular, we describe tools that automatically convert autonomy software into formal models that are then verified using model checking. This approach has been applied to MPL code for the Livingstone fault diagnosis system and to TDL task descriptions for mobile robot systems. Our long-term objective is to create tools that enable engineers and roboticists to use formal verification as part of the normal software development cycle.

This paper presents a general approach to robust execution monitoring. The goal is to provide coverage for many types of unexpected and unanticipated situations, while at the same time enabling the robot to quickly detect, and react to, specific contingencies. The approach uses a hierarchy of monitors, structured in layers of increasing specificity. We present the general approach, and show its application in the domain of indoor mobile robot navigation.

The lane-curvature method (LCM) presented in this paper is a new local obstacle avoidance method for indoor mobile robots. The method combines curvature-velocity method (CVM) with a new directional method called the lane method. The lane method divides the environment into lanes, and then chooses the best lane to follow to optimize travel along a desired heading. A local heading is then calculated for entering and following the best lane, and CVM uses this heading to determine the optimal translational and rotational velocities, considering the heading direction, physical limitations, and environmental constraints. By combining both the directional and velocity space methods, LCM yields safe collision-free motion as well as smooth motion taking the dynamics of the robot into account.

Robot systems must achieve high level goals while remaining reactive to contingencies and new opportunities. This typically requires robot systems to coordinate concurrent activities, monitor the environment, and deal with exceptions. We have developed a new language to support such task-level control. The language, TDL, is an extension of C++ that provides syntactic support for task decomposition, synchronization, execution monitoring, and exception handling. A compiler transforms TDL into pure C++ code that utilizes a platform-independent task management library. This paper introduces TDL, describes the task tree representation that underlies the language, and presents some aspects of its implementation and use in an autonomous mobile robot.

Navigation methods for office delivery robots need to take various sources of uncertainty into account in order to get robust performance. In previous work, we developed a reliable navigation technique that uses partially observable Markov models to represent metric, actuator and sensor uncertainties. This paper describes an algorithm that adjusts the probabilities of the initial Markov model by passively observing the robot's interactions with its environment. The learned probabilities more accurately reflect the actual uncertainties in the environment, which ultimately leads to improved navigation performance. The algorithm, an extension of the Baum-Welch algorithm, learns without a teacher and addresses the issues of limited memory and the cost of collecting training data. Empirical results show that the algorithm learns good Markov models with a small amount of training data.

We present a new method for local obstacle avoidance by indoor mobile robots that formulates the problem as one of constrained optimization in velocity space. Constraints that stem from physical limitations (velocities and accelerations) and the environment (the configuration of obstacles) are placed on the translational and rotational velocities of the robot. The robot chooses velocity commands that satisfy all the constraints and maximize an objective function that trades off speed, safety and goal-directedness. An efficient, real-time implementation of the method has been extensively tested, demonstrating reliable, smooth and speedy navigation in office environments. The obstacle avoidance method is used as the basis of more sophisticated navigation behaviors, ranging from simple wandering to map-based navigation.

Reliable navigation is critical for a lunar rover, both for autonomous traverses and safeguarded remote teleoperation. This paper describes an implemented system that has autonomously driven a prototype wheeled lunar rover over a kilometer in natural, outdoor terrain. The navigation system uses stereo terrain maps to perform local obstacle avoidance, and arbitrates steering recommendations from both the user and the rover. The paper describes the system architecture, each of the major components, and the experimental results to date.

To operate in rich, dynamic environments, autonomous robots must be able to effectively utilize and coordinate their limited physical and computational resources. As complexity increases, it becomes necessary to impose explicit constraints on the control of planning, perception, and action to ensure that unwanted interactions between behaviors do not occur. This paper advocates developing complex robot systems by layering reactive behaviors onto deliberative components. In this structured control approach, the deliberative components handle normal situations and the reactive behaviors, which are explicitly constrained as to when and how they are activated, handle exceptional situations. The Task Control Architecture (TCA) has been developed to support this approach. TCA provides an integrated set of control constructs useful for implementing deliberative and reactive behaviors. The control constructs facilitate modular and evolutionary system development: they are used to integrate and coordinate planning, perception, and execution, and to incrementally improve the efficiency and robustness of the robot systems. To date, TCA has been used in implementing a half-dozen mobile robot systems, including an autonomous six-legged rover and indoor mobile manipulator.>

The authors quantify several performance metrics for the Ambler, a six-legged robot configured for autonomous traversal of Mars-like terrain. They present power consumption measures for walking on sandy terrain and for vertical lifts at different velocities. They document the accuracy of a novel dead reckoning approach, and analyze the accuracy. They describe the results of autonomous walking experiments in terms of terrain traversed, walking speed, number of instructions executed and endurance.>

Robot systems require extreme self- reliance to successfully engage in long-term au- tonomous operation. This paper describes techniques developed to provide the necessary reliability for the Ambler, a six-legged robot designed for planetary ex- ploration. The approach utilizes the Task Control Architecture to incrementally add monitoring and er- ror recovery strategies without having to modify the basic walking competence of the robot. Deliberative and reactive behaviors are combined to deal with the challenges of efficiency and safety imposed by plane- tary missions. Experimental results of long-term au- tonomous walking in rough terrain are presented.

We propose a methodology to automatically generate plans for a robot excavator like a bucket loader or a backhoe. The task is formulated as one of constrained optimization in an actkm space that is spanned by the parameters of a prototypical digging plan. We show how geometric and force constraints are lmposed on the action space to build the set of feasible plans, and discuss methods to optimize a cost function within this set. We discuss recent simulation results that demonstrate this method in action.

As part of the planetary Rover project at Carnegie Mellon University, a system that autonomously navigates a legged robot through complex obstacle courses has been developed. The system is integrated using the task control architecture (TCA), which provides communication and coordination facilities. The walking system, as originally implemented, had a sequential sense-plan-act control cycle. Utilizing TCA features for task sequencing and monitoring, the system was modified to concurrently plan and execute steps. Overall walking speed increased significantly, with only a relatively modest conversion effort.>

The Carnegie Mellon University Planetary Rover project is developing the Ambler, a six-legged robot designed for planetary exploration. The authors have developed reliable and efficient control, perception and planning algorithms suitable for navigating rugged terrain. The components have been integrated into a system that autonomously walks the Ambler along routes and over obstacles.>