We introduce Refeeding State Embeddings aligned using Environmental Data (RESEED), a novel method for grounding language in environmental data.While large language models (LLMs) excel at many tasks, they continue to struggle with multi-step sequential reasoning.RESEED addresses this by producing latent embeddings aligned with the true state of the environment and refeeding these embeddings into the model before generating its output.To evaluate its effectiveness, we develop three new sequential reasoning benchmarks, each with a training set of paired state-text trajectories and several text-only evaluation sets that test generalization to longer trajectories.Across all benchmarks, RESEED significantly improves generalization and scalability over a text-only baseline.We further show that RE-SEED outperforms commercial LLMs on our benchmarks, highlighting the value of grounding language in the environment.1

Developing whole-body tactile skins for robots remains a challenging task, as existing solutions often prioritize modular, one-size-fits-all designs, which, while versatile, fail to account for the robot's specific shape and the unique demands of its operational context. In this work, we introduce GenTact Toolbox, a computational pipeline for creating versatile wholebody tactile skins tailored to both robot shape and application domain. Our method includes procedural mesh generation for conforming to a robot's topology, task-driven simulation to refine sensor distribution, and multi-material 3D printing for shape-agnostic fabrication. We validate our approach by creating and deploying six capacitive sensing skins on a Franka Research 3 robot arm in a human-robot interaction scenario. This work represents a shift from “one-size-fits-all” tactile sensors toward context-driven, highly adaptable designs that can be customized for a wide range of robotic systems and applications. The project website is available at https://hiro-group.ronc.one/gentacttoolbox

In collaborative tasks, autonomous agents fall short of humans in their capability to quickly adapt to new and unfamiliar teammates. We posit that a limiting factor for zero-shot coordination is the lack of shared task abstractions, a mechanism humans rely on to implicitly align with teammates. To address this gap, we introduce HA^2: Hierarchical Ad Hoc Agents, a framework leveraging hierarchical reinforcement learning to mimic the structured approach humans use in collaboration. We evaluate HA^2 in the Overcooked environment, demonstrating statistically significant improvement over existing baselines when paired with both unseen agents and humans, providing better resilience to environmental shifts, and outperforming all state-of-the-art methods.

Successful human-robot collaboration depends on cohesive communication and a precise understanding of the robot’s abilities, goals, and constraints. While robotic manipulators offer high precision, versatility, and productivity, they exhibit expressionless and monotonous motions that conceal the robot’s intention, resulting in a lack of efficiency and transparency with humans. In this work, we use Laban notation, a dance annotation language, to enable robotic manipulators to generate trajectories with functional expressivity, where the robot uses nonverbal cues to communicate its abilities and the likelihood of succeeding at its task. We achieve this by introducing two novel variants of Hesitant expressive motion (Spoke-Like and Arc-Like). We also enhance the emotional expressivity of four existing emotive trajectories (Happy, Sad, Shy, and Angry) by augmenting Laban Effort usage with Laban Shape. The functionally expressive motions are validated via a human-subjects study, where participants equate both variants of Hesitant motion with reduced robot competency. The enhanced emotive trajectories are shown to be viewed as distinct emotions using the Valence-Arousal-Dominance (VAD) spectrum, corroborating the usage of Laban Shape.

For safe and effective operation of humanoid robots in human-populated environments, the problem of commanding a large number of degrees of freedom (DoFs) while simultaneously considering dynamic obstacles and human proximity has still not been solved. In this article, we present a new reactive motion controller that commands two arms of a humanoid robot and three torso joints (17 DoF in total). We formulate a quadratic program that seeks joint velocity commands respecting multiple constraints while minimizing the magnitude of the velocities. We introduce a new unified treatment of obstacles that dynamically maps visual and proximity (precollision) and tactile (postcollision) obstacles as additional constraints to the motion controller, in a distributed fashion over the surface of the upper body of the iCub robot (with 2000 pressure-sensitive receptors). This results in a bioinspired controller that: first, gives rise to a robot with whole-body visuo-tactile awareness, resembling peripersonal space representations, and, second, produces human-like minimum jerk movement profiles. The controller was extensively experimentally validated, including a physical human–robot interaction scenario.

Understanding human intentions is critical for safe and effective human-robot collaboration. While state of the art methods for human goal prediction utilize learned models to account for the uncertainty of human motion data, that data is inherently stochastic and high variance, hindering those models' utility for interactions requiring coordination, including safety-critical or close-proximity tasks. Our key insight is that robot teammates can deliberately configure shared workspaces prior to interaction in order to reduce the variance in human motion, realizing classifier-agnostic improvements in goal prediction. In this work, we present an algorithmic approach for a robot to arrange physical objects and project "virtual obstacles'' using augmented reality in shared human-robot workspaces, optimizing for human legibility over a given set of tasks. We compare our approach against other workspace arrangement strategies using two human-subjects studies, one in a virtual 2D navigation domain and the other in a live tabletop manipulation domain involving a robotic manipulator arm. We evaluate the accuracy of human motion prediction models learned from each condition, demonstrating that our workspace optimization technique with virtual obstacles leads to higher robot prediction accuracy using less training data.

In this work, we introduce LazyBoE, a multi-query method for kinodynamic motion planning with forward propagation. This algorithm allows for the simultaneous exploration of a robot’s state and control spaces, thereby enabling a wider suite of dynamic tasks in real-world applications. Our contributions are three-fold: i) a method for discretizing the state and control spaces to amortize planning times across multiple queries; ii) lazy approaches to collision checking and propagation of control sequences that decrease the cost of physics-based simulation; and iii) LazyBoE, a robust kinodynamic planner that leverages these two contributions to produce dynamically-feasible trajectories. The proposed framework not only reduces planning time but also increases success rate in comparison to previous approaches.

In this work, we present a novel algorithm to perform spill-free handling of open-top liquid-filled containers that operates in cluttered environments. By allowing liquid-filled containers to be tilted at higher angles and enabling motion along all axes of end-effector orientation, our work extends the reachable space and enhances maneuverability around obstacles, broadening the range of feasible scenarios. Our key contributions include: i) generating spill-free paths through the use of RRT* with an informed sampler that leverages container properties to avoid spill-inducing states (such as an upside-down container), ii) parameterizing the resulting path to generate spill-free trajectories through the implementation of a time parameterization algorithm, coupled with a transformer-based machine-learning model capable of classifying trajectories as spill-free or not. We validate our approach in real-world, obstacle-rich task settings using containers of various shapes and fill levels and demonstrate an extended solution space that is at least 3x larger than an existing approach.

This work explores non-prehensile manipulation (NPM) and whole-body interaction as strategies for enabling robotic manipulators to conduct manipulation tasks despite experiencing locked multi-joint (LMJ) failures. LMJs are critical system faults where two or more joints become inoperable; they impose constraints on the robot’s configuration and control spaces, consequently limiting the capability and reach of a prehensile-only approach. This approach involves three components: i) modeling the failure-constrained workspace of the robot, ii) generating a kinodynamic map of NPM actions within this workspace, and iii) a manipulation action planner that uses a sim-in-the-loop approach to select the best actions to take from the kinodynamic map. The experimental evaluation shows that our approach can increase the failure-constrained reachable area in LMJ cases by 79%. Further, it demonstrates the ability to complete real-world manipulation with up to 88.9% success when the end-effector is unusable and up to 100% success when it is usable.

Effective collaboration between humans and AIs hinges on transparent communication and alignment of mental models. However, explicit, verbal communication is not always feasible. Under such circumstances, human-human teams often depend on implicit, nonverbal cues to glean important information about their teammates such as intent and expertise, thereby bolstering team alignment and adaptability. Among these implicit cues, two of the most salient and fundamental are a human’s actions in the environment and their visual attention. In this paper, we present a novel method to combine eye gaze data and behavioral data, and evaluate their respective predictive power for human proficiency, trust, and intent. We first collect a dataset of paired eye gaze and gameplay data in the fast-paced collaborative "Overcooked" environment. We then train models on this dataset to compare how the predictive powers differ between gaze data, gameplay data, and their combination. We additionally compare our method to prior works that aggregate eye gaze data and demonstrate how these aggregation methods can substantially reduce the predictive ability of eye gaze. Our results indicate that, while eye gaze data and gameplay data excel in different situations, a model that integrates both types consistently outperforms all baselines. This work paves the way for developing intuitive and responsive agents that can efficiently adapt to new teammates.

For humans to effectively work with robots, they must be able to predict the actions and behaviors of their robot teammates rather than merely react to them. While there are existing techniques enabling robots to adapt to human behavior, there is a demonstrated need for methods that explicitly improve humans’ ability to understand and predict robot behavior at multi-task timescales. In this work, we propose a method leveraging the innate human propensity for pattern recognition in order to improve team dynamics in human–robot teams and to make robots more predictable to the humans that work with them. Patterns are a cognitive tool that humans use and rely on often, and the human brain is in many ways primed for pattern recognition and usage. We propose pattern-aware convention-setting for teaming (PACT), an entropy-based algorithm that identifies and imposes appropriate patterns over a robot’s planner or policy over long time horizons. These patterns are autonomously generated and chosen via an algorithmic process that considers human-perceptible features and characteristics derived from the tasks to be completed, and as such, produces behavior that is easier for humans to identify and predict. Our evaluation shows that PACT contributes to significant improvements in team dynamics and teammate perceptions of the robot, as compared to robots that utilize traditionally ‘optimal’ plans and robots utilizing unoptimized patterns.

One important aspect of effective human--robot collaborations is the ability for robots to adapt quickly to the needs of humans. While techniques like deep reinforcement learning have demonstrated success as sophisticated tools for learning robot policies, the fluency of human-robot collaborations is often limited by these policies' inability to integrate changes to a user's preferences for the task. To address these shortcomings, we propose a novel approach that can modify learned policies at execution time via symbolic if-this-then-that rules corresponding to a modular and superimposable set of low-level constraints on the robot's policy. These rules, which we call Transparent Matrix Overlays, function not only as succinct and explainable descriptions of the robot's current strategy but also as an interface by which a human collaborator can easily alter a robot's policy via verbal commands. We demonstrate the efficacy of this approach on a series of proof-of-concept cooking tasks performed in simulation and on a physical robot.

Current motion planning approaches rely on binary collision checking to evaluate the validity of a state and thereby dictate where the robot is allowed to move. This approach leaves little room for robots to engage in contact with an object, as is often necessary when operating in densely cluttered spaces. In this work, we propose an alternative method that considers contact states as high-cost states that the robot should avoid but can traverse if necessary to complete a task. More specifically, we introduce Contact Admissible Transition-based Rapidly exploring Random Trees (CAT-RRT)11Supplementary video and open source code [1]., a planner that uses a novel per-link cost heuristic to find a path by traversing high-cost obstacle regions. Through extensive testing, we find that state-of-the-art optimization planners tend to over-explore low-cost states, which leads to slow and inefficient convergence to contact regions. Conversely, CAT-RRT searches both low and high-cost regions simultaneously with an adaptive thresholding mechanism carried out at each robot link. This leads to paths with a balance between efficiency, path length, and contact cost.

Real-time control is an essential aspect of safe robot operation in the real world with dynamic objects. We present a framework for the analysis of object-aware con-trollers, methods for altering a robot's motion to anticipate and avoid possible collisions. This framework is focused on three design considerations: kinematics, motion profiles, and virtual constraints. Additionally, the analysis in this work relies on verification of robot behaviors using fundamental robot-obstacle experimental scenarios. To showcase the effectiveness of our method we compare three representative object-aware controllers. The comparison uses metrics originating from the design considerations. From the analysis, we find that the design of object-aware controllers often lacks kinematic considerations, continuity of control points, and stability in movement profiles. We conclude that this framework can be used in the future to design, compare, and benchmark obstacle avoidance methods.

Kitting refers to the task of preparing and grouping necessary parts and tools (or “kits”) for assembly in a manufacturing environment. Automating this process simplifies the assembly task for human workers and improves efficiency. Existing automated kitting systems adhere to scripted instructions and predefined heuristics. However, given variability in the availability of parts and logistic delays, the inflexibility of existing systems can limit the overall efficiency of an assembly line. In this paper, we propose a bilevel optimization framework to enable a robot to perform task segmentation-based part selection, kit arrangement, and delivery scheduling to provide custom-tailored kits just in time—i.e., right when they are needed. We evaluate the proposed approach both through a human subjects study (n=18) involving the construction of a flat-pack furniture table and shop-flow simulation based on the data from the study. Our results show that the just-in-time kitting system is objectively more efficient, resilient to upstream shop flow delays, and subjectively more preferable as compared to baseline approaches of using kits defined by rigid task segmentation boundaries defined by the task graph itself or a single kit that includes all parts necessary to assemble a single unit.

Recent work has raised concerns about the inherent limitations of text-only pretraining.In this paper, we first demonstrate that reporting bias, the tendency of people to not state the obvious, is one of the causes of this limitation, and then investigate to what extent multimodal training can mitigate this issue.To accomplish this, we 1) generate the Color Dataset (CoDa), a dataset of human-perceived color distributions for 521 common objects; 2) use CoDa to analyze and compare the color distribution found in text, the distribution captured by language models, and a human's perception of color; and 3) investigate the performance differences between text-only and multimodal models on CoDa.Our results show that the distribution of colors that a language model recovers correlates more strongly with the inaccurate distribution found in text than with the ground-truth, supporting the claim that reporting bias negatively impacts and inherently limits text-only training.We then demonstrate that multimodal models can leverage their visual training to mitigate these effects, providing a promising avenue for future research.* *Email has no accent, but includes the hyphen. 1 In this paper, we use LM to refer to both causal LMs as well as masked LMs.

In this paper, we present a framework that unites obstacle avoidance and deliberate physical interaction for robotic manipulators. As humans and robots begin to coexist in work and household environments, pure collision avoidance is insufficient, as human–robot contact is inevitable and, in some situations, desired. Our work enables manipulators to anticipate, detect, and act on contact. To achieve this, we allow limited deviation from the robot’s original trajectory through velocity reduction and motion restrictions. Then, if contact occurs, a robot can detect it and maneuver based on a novel dynamic contact thresholding algorithm. The core contribution of this work is dynamic contact thresholding, which allows a manipulator with onboard proximity sensors to track nearby objects and reduce contact forces in anticipation of a collision. Our framework elicits natural behavior during physical human–robot interaction. We evaluate our system on a variety of scenarios using the Franka Emika Panda robot arm; collectively, our results demonstrate that our contribution is not only able to avoid and react on contact, but also anticipate it.

In this paper, we present an accelerometer-based kinematic calibration algorithm to accurately estimate the pose of multiple sensor units distributed along a robot body. Our approach is self-contained, can be used on any robot provided with a Denavit-Hartenberg kinematic model, and on any skin equipped with Inertial Measurement Units (IMUs). To validate the proposed method, we first conduct extensive experimentation in simulation and demonstrate a sub-cm positional error from ground truth data—an improvement of six times with respect to prior work; subsequently, we then perform a real-world evaluation on a seven degrees-of-freedom collaborative platform. For this purpose, we additionally introduce a novel design for a stand-alone artificial skin equipped with an IMU for use with the proposed algorithm and a proximity sensor for sensing distance to nearby objects. In conclusion, in this work, we demonstrate seamless integration between a novel hardware design, an accurate calibration method, and preliminary work on applications: the high positional accuracy effectively enables to locate distributed proximity data and allows for a distributed avoidance controller to safely avoid obstacles and people without the need of additional sensing.

Multi-robot cooperation requires agents to make decisions that are consistent with the shared goal without disregarding action-specific preferences that might arise from asymmetry in capabilities and individual objectives. To accomplish this goal, we propose a method named SLiCC: Stackelberg Learning in Cooperative Control. SLiCC models the problem as a partially observable stochastic game composed of Stackelberg bimatrix games, and uses deep reinforcement learning to obtain the payoff matrices associated with these games. Appropriate cooperative actions are then selected with the derived Stackelberg equilibria. Using a bi-robot cooperative object transportation problem, we validate the performance of SLiCC against centralized multi-agent Q-learning and demonstrate that SLiCC achieves better combined utility.

With robots leaving factories and entering less controlled domains, possibly sharing the space with humans, safety is paramount and multimodal awareness of the body surface and the surrounding environment is fundamental. Taking inspiration from peripersonal space representations in humans, we present a framework on a humanoid robot that dynamically maintains such a protective safety zone, composed of the following main components: (i) a human 2D keypoints estimation pipeline employing a deep learning based algorithm, extended here into 3D using disparity; (ii) a distributed peripersonal space representation around the robot»s body parts; (iii) a reaching controller that incorporates all obstacles entering the robot»s safety zone on the fly into the task. Pilot experiments demonstrate that an effective safety margin between the robot»s and the human»s body parts is kept. The proposed solution is flexible and versatile since the safety zone around individual robot and human body parts can be selectively modulated---here we demonstrate stronger avoidance of the human head compared to rest of the body. Our system works in real time and is self-contained, with no external sensory equipment and use of onboard cameras only.

Research in human teamwork shows that a key element of fluid and fluent interactions is the interpretation of implicit verbal and non-verbal cues in context. This poses an issue to robotic platforms, however, as they have historically worked best when controlled through explicit commands that have employed structured, unequivocal representations of the external world and their human partners. In this work, we present a framework for effectively grounding situated and naturalistic speech to action selection during human-robot collaborative activities. This is accomplished by maintaining and incrementally updating separate “speech” and “context” models that jointly classify a collaborator's utterance. We evaluate the efficacy of the system on a collaborative construction task with an autonomous robot and human participants. We first demonstrate that our system is capable of acquiring and deploying new task representations from limited and naturalistic data sets, and without any prior domain knowledge of language or the task itself. Finally, we show that our system is capable of significantly improving performance on an unfamiliar task after a one-shot exposure.

Human-Robot Collaboration (HRC) aims to develop robots that provide assistance to human workers while performing physical tasks. Such assistance comes in the form of supportive behaviors that are different from the actions part of the task, and that are meant to help a human worker more effectively accomplish the task. Learning how to provide useful behaviors that are tailored to a human peer represents a difficult challenge. This is due to the need of large amounts of training data in the form of real world observations that include information about such preferences. This data needs to encode not only the structure and progression of the task, but also the different workers' preferences with respect to when and what assistance the robot should provide. Our work separates the challenge of learning a model of the task (which requires a large amount of training data) from that of learning supportive behavior preferences for the interaction (which has obvious restrictions for the number of user-provided demonstrations to which we have access). We first learn a hidden Markov model (HMM) from a training set consisting of observed human workers performing the considered task in simulation. We then use this model to predict, while observing the human peer, what supportive behaviors a robot should offer throughout the task. Building upon the hidden state representation, our system is able to learn the supportive behaviors based on as few as five user-annotated demonstrations, learning a personalized supportive behavior model. We evaluate our system on a user study with 14 participants, and show results on par with human-level prediction for the task.

In this paper, we present a model set for designing human-robot collaboration (HRC) experiments. It targets a common scenario in HRC, which is the collaborative assembly of furniture, and it consists of a combination of standard components and custom designs. With this work, we aim at reducing the amount of work required to set up and reproduce HRC experiments, and we provide a unified framework to facilitate the comparison and integration of contributions to the field. The model set is designed to be modular, extendable, and easy to distribute. Importantly, it covers the majority of relevant research in HRC, and it allows tuning of a number of experimental variables that are particularly valuable to the field. Additionally, we provide a set of software libraries for perception, control and interaction, with the goal of encouraging other researchers to proactively contribute to our work.

Collaborative robots represent a clear added value to manufacturing, as they promise to increase productivity and improve working conditions of such environments. Although modern robotic systems have become safe and reliable enough to operate close to human workers on a day-to-day basis, the workload is still skewed in favor of a limited contribution from the robot's side, and a significant cognitive load is allotted to the human. We believe the transition from robots as recipients of human instruction to robots as capable collaborators hinges around the implementation of transparent systems, where mental models about the task are shared between peers, and the human partner is freed from the responsibility of taking care of both actors. In this work, we implement a transparent task planner able to be deployed in realistic, near-future applications. The proposed framework is capable of basic reasoning capabilities for what concerns role assignment and task allocation, and it interfaces with the human partner at the level of abstraction he is most comfortable with. The system is readily available to non-expert users, and programmable with high-level commands in an intuitive interface. Our results demonstrate an overall improvement in terms of completion time, as well as a reduced cognitive load for the human partner.

With robots leaving factory environments and entering less controlled domains, possibly sharing living space with humans, safety needs to be guaranteed. To this end, some form of awareness of their body surface and the space surrounding it is desirable. In this work, we present a unique method that lets a robot learn a distributed representation of space around its body (or peripersonal space) by exploiting a whole-body artificial skin and through physical contact with the environment. Every taxel (tactile element) has a visual receptive field anchored to it. Starting from an initially blank state, the distance of every object entering this receptive field is visually perceived and recorded, together with information whether the object has eventually contacted the particular skin area or not. This gives rise to a set of probabilities that are updated incrementally and that carry information about the likelihood of particular events in the environment contacting a particular set of taxels. The learned representation naturally serves the purpose of predicting contacts with the whole body of the robot, which is of clear behavioral relevance. Furthermore, we devised a simple avoidance controller that is triggered by this representation, thus endowing a robot with a “margin of safety” around its body. Finally, simply reversing the sign in the controller we used gives rise to simple “reaching” for objects in the robot's vicinity, which automatically proceeds with the most activated (closest) body part.

Calibration continues to receive significant attention in robotics because of its key impact on performance and cost associated with the operation of complex robots. Calibration of kinematic parameters is typically the first mandatory step. To this end, a variety of metrology systems and corresponding algorithms have been described in the literature relying on measurements of the pose of the end-effector using a camera or laser tracking system, or, exploiting constraints arising from contacts of the end-effector with the environment. In this work, we take inspiration from the behavior of infants and certain animals, who are believed to use self-stimulation or self-touch to “calibrate” their body representations, and present a new solution to this problem by letting the robot close the kinematic chain by touching its own body. The robot considered in this paper is sensorized with tactile arrays for a total of about 4200 sensing points. The correspondence between the predicted contact point from existing forward kinematics and the actual position on the robot's `skin' provides sample data that allows refining the kinematic representation (DH parameters). The data collection procedure is automated - self-touch is autonomously executed by the robot - and can be repeated at any time, providing a compact self-calibration system that does not require an external measurement apparatus.