Personalized medication plans determine the selection, dosage, and administration schedule of medications, to achieve medical goals that are specific to the patient and to its individual health constraints. This paper introduces medication planning as a novel domain for artificial intelligence planning, using PDDL+. We evaluate the suggested representation via experiments based on data collected from medical studies conducted on mice and rats.

The shortest path problem in graphs is fundamental to AI. Nearly all variants of the problem and relevant algorithms that solve them ignore edge-weight computation time and its common relation to weight uncertainty. This implies that taking these factors into consideration can potentially lead to a performance boost in relevant applications. Recently, a generalized framework for weighted directed graphs was suggested, where edge-weight can be computed (estimated) multiple times, at increasing accuracy and run-time expense. We build on this framework to introduce the problem of finding the tightest admissible shortest path (TASP); a path with the tightest suboptimality bound on the optimal cost. This is a generalization of the shortest path problem to bounded uncertainty, where edge-weight uncertainty can be traded for computational cost. We present a complete algorithm for solving TASP, with guarantees on solution quality. Empirical evaluation supports the effectiveness of this approach.

Naturally occurring collective motion is a fascinating phenomenon in which swarming individuals aggregate and coordinate their motion. Many theoretical models of swarming assume idealized, perfect perceptual capabilities, and ignore the underlying perception processes, particularly for agents relying on visual perception. Specifically, biological vision in many swarming animals, such as locusts, utilizes monocular non-stereoscopic vision, which prevents perfect acquisition of distances and velocities. Moreover, swarming peers can visually occlude each other, further introducing estimation errors. In this study, we explore necessary conditions for the emergence of ordered collective motion under restricted conditions, using non-stereoscopic, monocular vision. We present a model of vision-based collective motion for locust-like agents: elongated shape, omni-directional visual sensor parallel to the horizontal plane, and lacking stereoscopic depth perception. The model addresses (i) the non-stereoscopic estimation of distance and velocity, (ii) the presence of occlusions in the visual field. We consider and compare three strategies that an agent may use to interpret partially-occluded visual information at the cost of the computational complexity required for the visual perception processes. Computer-simulated experiments conducted in various geometrical environments (toroidal, corridor, and ring-shaped arenas) demonstrate that the models can result in an ordered or near-ordered state. At the same time, they differ in the rate at which order is achieved. Moreover, the results are sensitive to the elongation of the agents. Experiments in geometrically constrained environments reveal differences between the models and elucidate possible tradeoffs in using them to control swarming agents. These suggest avenues for further study in biology and robotics.

The shortest path problem in graphs is a cornerstone of AI theory and applications. Existing algorithms generally ignore edge weight computation time. We present a generalized framework for weighted directed graphs, where edge weight can be computed (estimated) multiple times, at increasing accuracy and run-time expense. This raises several generalized variants of the shortest path problem. We introduce the problem of finding a path with the tightest lower-bound on the optimal cost. We then present two complete algorithms for the generalized problem, and empirically demonstrate their efficacy.

The extraction of features is an essential step in the process of mining software repositories. An important feature that has been actively studied in the field of mining software repositories is bad code smells. Bad code smells are patterns in the source code that indicate an underlying issue in the design and implementation of the software. Several tools have been proposed to extract code smells. However, currently, there are no tools that extract a significant number of code smells from software written in C++. Therefore, we propose CLEAN++ (Code smeLls ExtrActioN for C++) [1]. It is an extension of a robust static code analysis tool that implements 35 code smells. To evaluate CLEAN++, we ran it over 44 open-source projects and wrote test cases to validate each code smell. Also, we converted the test cases to Java and used two Java tools to validate the effectiveness of our tool. In the end, we confirmed that the CLEAN++ is successful at detecting code smells.The tool is available at https://github.com/Tomma94/CLEAN-Plus-Plus.

Plan recognition is the task of inferring the plan of an agent, based on an incomplete sequence of its observed actions. Previous formulations of plan recognition commit early to discretizations of the environment and the observed agent's actions. This leads to reduced recognition accuracy. To address this, we first provide a formalization of recognition problems which admits continuous environments, as well as discrete domains. We then show that through mirroring---generalizing plan-recognition by planning---we can apply continuous-world motion planners in plan recognition. We provide formal arguments for the usefulness of mirroring, and empirically evaluate mirroring in more than a thousand recognition problems in three continuous domains and six classical planning domains.

Models of crowd dynamics are critically important for urban planning and management. They support analysis, facilitate qualitative and quantitative predictions, and synthesize behaviors for simulations. One promising approach to crowd modeling relies on micro-level agent-based simulations, where the interactions of simulated individual agents in the crowd result in macro-level crowd dynamics which are the object of study. This article reports on an agent-based model of urban pedestrian crowds, where culture is explicitly modeled . We extend an established agent-based social agent model, inspired by social psychology, to account for individual cultural attributes discussed in social science literature. We then embed the model in a simulation of pedestrians and explore the resulting macro-level crowd behaviors, such as pedestrian flow, lane changes rate, and so on. We validate the model by quantitatively comparing the simulation results to the pedestrian dynamics in movies of human crowds in five different countries: Iraq, Israel, England, Canada, and France. We conclude that the model can faithfully replicate urban pedestrians in different cultures. Encouraged by these results, we explore simulations of mixed-culture pedestrian crowds.

Goal recognition is the problem of inferring the goal of an agent, based on its observed actions. An inspiring approach—plan recognition by planning (PRP)—uses off-the-shelf planners to dynamically generate plans for given goals, eliminating the need for the traditional plan library. However, existing PRP formulation is inherently inefficient in online recognition, and cannot be used with motion planners for continuous spaces. In this paper, we utilize a different PRP formulation which allows for online goal recognition, and for application in continuous spaces. We present an online recognition algorithm, where two heuristic decision points may be used to improve run-time significantly over existing work. We specify heuristics for continuous domains, prove guarantees on their use, and empirically evaluate the algorithm over hundreds of experiments in both a 3D navigational environment and a cooperative robotic team task.

Asimov's three laws of robotics, which were shaped in the literary work of Isaac Asimov (1920-1992) and others, define a crucial code of behavior that fictional autonomous robots must obey as a condition for their integration into human society. While, general implementation of these laws in robots is widely considered impractical, limited-scope versions have been demonstrated and have proven useful in spurring scientific debate on aspects of safety and autonomy in robots and intelligent systems. In this work, we use Asimov's laws to examine these notions in molecular robots fabricated from DNA origami. We successfully programmed these robots to obey, by means of interactions between individual robots in a large population, an appropriately scoped variant of Asimov's laws, and even emulate the key scenario from Asimov's story "Runaround," in which a fictional robot gets into trouble despite adhering to the laws. Our findings show that abstract, complex notions can be encoded and implemented at the molecular scale, when we understand robots on this scale on the basis of their interactions.

Human operators play a key role in robotic exploration and search missions, as the interpretation of camera images typically requires the visual perception skills of humans. Thus one the key challenges in building effective robotic systems for such missions lies in developing good operator interfaces. In this paper, we present a novel asynchronous user-guided interface for human operators of robotic search of an unknown area. Enabled by efficient methods to store and retrieve recorded images (and meta information) in real-time, our interface allows the operator to click on any point of interest. The operator is then presented with highly-relevant images that cover the point, without occlusion. This, in contrast with system-guided approaches, where an automated system selects areas and images for inspection. Experiments with 32 human subjects in two different-size maps favor the user-guided approach we present over the system-guided approach. Additional experiments with human subjects provide an explanation as to the environment characteristics that favor our approach.

Coverage is a fundamental problem in robotics, where one or more robots are required to visit each point in a target area at least once. While most previous work concentrated on finding a solution that completes the coverage as quickly as possible, in this paper we consider a new version of the problem: adversarial coverage. Here, the robot operates in an environment that contains threats that might stop the robot. We introduce the problem of finding the safest adversarial coverage path, and present different optimization criteria for the evaluation of these paths. We show that finding an optimal solution to the safest coverage problem is NP-Complete. We therefore suggest two heuristic algorithms: STAC, a spanning-tree based coverage algorithm, and GSAC, which follows a greedy approach. These algorithms produce close to optimal solutions in polynomial time. We establish theoretical bounds on the total risk involved in the coverage paths created by these algorithms and on their lengths. Lastly, we compare the effectiveness of these two algorithms in various types of environments and settings.

This paper discusses the problem of generating efficient coverage paths for a mobile robot in an adversarial environment, where threats exist that might stop the robot. First, we formally define the problem of adversarial coverage, and present optimization criteria used for evaluation of coverage algorithms in adversarial environments. We then present a coverage area planning algorithm based on a map of the probable threats. The algorithm tries to minimize the total risk involved in covering the target area while taking into account coverage time constrains. The algorithm is based on incrementally extending the coverage path to the nearest safe cells while allowing the robot to repeat its steps. By allowing the robot to visit each cell in the target area more than once, the accumulated risk can be reduced at the expense of extending the coverage time. We show the effectiveness of this algorithm in extensive experiments.

The ability to model and reason about the potential violence level of a demonstration is important to the police decision making process. Unfortunately, existing knowledge regarding demonstrations is composed of partial qualitative descriptions without complete and precise numerical information. In this article we describe a first attempt to use qualitative reasoning techniques to model demonstrations. To our knowledge, such techniques have never been applied to modeling and reasoning regarding crowd behaviors, nor in particular demonstrations. We develop qualitative models consistent with the partial, qualitative social science literature, allowing us to model the interactions between different factors that influence violence in demonstrations. We then utilize qualitative simulation to predict the potential eruption of violence, at various levels, based on a description of the demographics, environmental settings, and police responses. We incrementally present and compare three such qualitative models. The results show that while two of these models fail to predict the outcomes of real-world events reported and analyzed in the literature, one model provides good results. We also examine whether a popular machine learning algorithm (decision tree learning) can be used. While the results show that the decision trees provide improved predictions, we show that the QR models can be more sensitive to changes, and can account for what-if scenarios, in contrast to decision trees. Moreover, we introduce a novel analysis algorithm that analyzes the QR simulations, to automatically determine the factors that are most important in influencing the outcome in specific real-world demonstrations. We show that the algorithm identifies factors that correspond to experts' analysis of these events.

Agents in a team should be in agreement. Unfortunately, they may come to disagree due to sensor uncertainty, intermittent communication failures, etc. Once a disagreement occurs, the agents should detect and diagnose the disagreement. Current diagnostic techniques do not scale well with the number of agents, as they have high communication and computation complexity. We present novel techniques that enable scalability in three ways. First, we use communications early in the diagnostic process to stave off unneeded reasoning, which ultimately leads to unneeded communications. Second, we use light‐weight (and inaccurate) behavior recognition to focus the diagnostic reasoning on beliefs of agents that might be in conflict. Finally, we propose diagnosing only to a limited number of representative agents (instead of all the agents). We examine these techniques in large‐scale teams of situated agents in two domains and show that combining the techniques produces a diagnostic process that is highly scalable in both communication and computation.

To make good decisions in a social context, humans often need to recognize the plan underlying the behavior of others, and make predictions based on this recognition. This process, when carried out by software agents or robots, is known as plan recog

As autonomous agents proliferate in the real world, both in software and robotic settings, they will increasingly need to band together for cooperative activities with previously unfamiliar teammates. In such ad hoc team settings, team strategies cannot be developed a priori. Rather, an agent must be prepared to cooperate with many types of teammates: it must collaborate without pre-coordination. This paper challenges the AI community to develop theory and to implement prototypes of ad hoc team agents. It defines the concept of ad hoc team agents, specifies an evaluation paradigm, and provides examples of possible theoretical and empirical approaches to challenge. The goal is to encourage progress towards this ambitious, newly realistic, and increasingly important research goal.

Multi-robot systems researchers have been investigating adaptive coordination methods for improving spatial coordination in teams. Such methods adapt the coordination method to the dynamic changes in density of the robots. Unfortunately, while their empirical success is evident, none of these methods has been understood in the context of existing formal work on multi-robot learning. This paper presents a reinforcement-learning approach to coordination algorithm selection, which is not only shown to work well in experiments, but is also analytically grounded. We present a reward function (Effectiveness Index, EI), that reduces time and resources spent coordinating, and maximizes the time between conflicts that require coordination. It does this by measuring the resource-spending velocity. We empirically show its success in simulations of multi-robot foraging. In addition, we analytically explore the reasons that EI works well. We show that under some assumptions, spatial coordination opportunities can be modeled as matrix games in which the payoffs are directly a function of EI estimates. The use of reinforcement learning leads to robots maximizing their EI rewards in equilibrium. This work is a step towards bridging the gap between the theoretical study of interactions, and their use in multi-robot coordination.

The use of unmanned autonomous vehicles is becoming more and more significant in recent years. The fact that the vehicles are unmanned (whether autonomous or not), can lead to greater difficulties in identifying failure and anomalous states, since the operator cannot rely on its own body perceptions to identify failures. Moreover, as the autonomy of unmanned vehicles increases, it becomes more difficult for operators to monitor them closely, and this further exacerbates the difficulty of identifying anomalous states, in a timely manner. Model-based diagnosis and fault-detection systems have been proposed to recognize failures. However, these rely on the capabilities of the underlying model, which necessarily abstracts away from the physical reality of the robot. In this paper we propose a novel, model-free, approach for detecting anomalies in unmanned autonomous vehicles, based on their sensor readings (internal and external). Experiments conducted on Unmanned Aerial Vehicles (UAVs) and Unmanned Ground Vehicles (UGVs) demonstrate the efficacy of the approach by detecting the vehicles deviations from nominal behavior.

We present an approach for learning models that obtain accurate classification of large scale data objects, collected in spatiotemporal domains. The model generation is structured in three phases: pixel selection (spatial dimension reduction), spatiotemporal features extraction and feature selection. Novel techniques for the first two phases are presented, with two alternatives for the middle phase. Model generation based on the combinations of techniques from each phase is explored. The introduced methodology is applied on datasets from the Voltage-Sensitive Dye Imaging (VSDI) domain, where the generated classification models successfully decode neuronal population responses in the visual cortex of behaving animals. VSDI currently is the best technique enabling simultaneous high spatial (10,000 points) and temporal (10 ms or less) resolution imaging from neuronal population in the cortex. We demonstrate that not only our approach is scalable enough to handle computationally challenging data, but it also contributes to the neuroimaging field of study with its decoding abilities.

In this paper, we investigate the use of hidden semi-Markov models (HSMMs) in analyzing data of human activities, a task commonly referred to as activity recognition. In particular, we use the models to recognize normal and abnormal two-dimensional j

Proceeding of: European Conference on Artificial Intelligence (ECAI 2008). Patras, Greece, july, 21st-25th, 2008.

This paper considers the problem of multi-robot patrol around a closed area with the existence of an adversary attempting to penetrate into the area. In case the adversary knows the patrol scheme of the robots and the robots use a deterministic patrol algorithm, then in many cases it is possible to penetrate with probability 1. Therefore this paper considers a non-deterministic patrol scheme for the robots, such that their movement is characterized by a probability p. This patrol scheme allows reducing the probability of penetration, even under an assumption of a strong opponent that knows the patrol scheme. We offer an optimal polynomial-time algorithm for finding the probability p such that the minimal probability of penetration detection throughout the perimeter is maximized. We describe three robotic motion models, defined by the movement characteristics of the robots. The algorithm described herein is suitable for all three models.

In formation-maintenance (formation control) tasks, robots maintain their relative position with respect to their peers, according to a desired geometric shape. Previous work has examined formation-maintenance algorithms, based on formation control graphs, that ensure the theoretical stability of the formation. However, an exponential number of stable controllers exists. Thus a key question is how to select (construct) a formation controller that optimizes desired properties, such as sensor usage. We present a novel representation of the sensing capabilities of robots in formations, using a monitoring multigraph. We first show that graph-theoretic techniques can then be used to efficiently compute optimal sensing policies that maintain a given formation, while minimizing sensing costs. In particular, separation-bearing (distance-angle) control targets are automatically constructed for each individual robot in the formation, taking into account its specific sensor morphology. Then, we present a protocol allowing control graphs to be switched on line, to allow robots to adjust to sensory failures. We report on results from comprehensive experiments with physical and simulated robots. The results show that the use of the dynamic protocol allows formations of real robots to move significantly faster and with greater precision, while reducing the number of formation failures, due to sensor limitations. We also evaluate the sensitivity of our approach to communication reliability, and discuss opportunities and challenges raised by our approach.

This paper discusses the problem of generating patrol paths for a team of mobile robots inside a designated target area. Patrolling requires an area to be visited repeatedly by the robot(s) in order to monitor its current state. First, we present frequency optimization criteria used for evaluation of patrol algorithms. We then present a patrol algorithm that guarantees maximal uniform frequency, i.e., each point in the target area is covered at the same optimal frequency. This solution is based on finding a circular path that visits all points in the area, while taking into account terrain directionality and velocity constraints. Robots are positioned uniformly along this path, using a second algorithm. Moreover, the solution is guaranteed to be robust in the sense that uniform frequency of the patrol is achieved as long as at least one robot works properly.

Recent years are seeing a renewed interest in general multi-robot architectures, capable of automating coordination. However, few architectures explore integration of multiple coordination mechanisms. Thus the question of how to best integrate coordination mechanisms is left open. This paper focuses on the micro-kernel integration approach used in BITE (Bar Ilan Teamwork Engine), a multi-robot behavior-based architecture. This approach allows the developer to plug in coordination mechanisms (teamwork behaviors) to be used depending on the context of execution. BITE imposes constraints on the specification of taskwork behaviors, which allow BITE's control algorithm to automatically determine sequencing and task-allocation points during task execution. At such points, teamwork behaviors (known as interaction behaviors) are triggered to automate the coordination processes. We argue that BITE's approach is preferable to the methodology of existing architectures, and provide analysis of experiments in using BITE with Sony AIBO robots, to support our argument

This paper discusses the problem of building efficient coverage paths for a team of robots. An efficient multirobot coverage algorithm should result in a coverage path for every robot, such that the union of all paths generates a full coverage of the terrain and the total coverage time is minimized. A method, underlying several coverage algorithms, suggests the use of spanning trees as base for creating coverage paths. Current studies assume that the spanning tree is given, and try to make the most out of the given configuration. However, overall performance of the coverage is heavily dependent on the given spanning tree. This paper tackles the open challenge of constructing a coverage spanning tree that minimizes the time to complete coverage. We argue that the choice of the initial spanning tree has far reaching consequences concerning the coverage time, and if the tree is constructed appropriately, it could considerably reduce the coverage time of the terrain. Therefore the problem studied here is finding spanning trees that would decrease the coverage time of the terrain when used as base for multi-robot coverage algorithms. The main contributions of this paper are twofold. First, it provides initial sound discussion and results concerning the construction of the tree as a crucial base for any efficient coverage algorithm. Second, it describes a polynomial-time tree construction algorithm that, as shown in extensive simulations, dramatically improves the coverage time even when used as a basis for a simple, inefficient, coverage algorithm

Area coverage is an important task for mobile robots, with many real-world applications. In many cases, the coverage has to be completed without the use of a map or any a priori knowledge about the area, a process referred-to as on-line coverage. Previous investigations of multi-robot on-line coverage focused on the improved efficiency gained from the use of multiple robots, but did not formally addressed the potential for greater robustness. We present a novel multi-robot on-line coverage algorithm, based on approximate cell decomposition. We analytically show that the algorithm is complete and robust, in that as long as a single robot is able to move, the coverage would be completed. We analyze the assumptions underlying the algorithm requirements and present a number of techniques for executing it in real robots. We show empirical coverage-time results of running the algorithm in two different environments and several group sizes

Many robotics applications require a human operator to monitor multiple robots that collaborate to achieve the operator's goals. Most approaches to such monitoring focus on each robot independently of its peers. When robots are tightly-coordinated, the operator is thus cognitively burdened to build a mental picture of the state of coordination. We report on extensive experiments (approximately 100 hours) with up to 25 human operators, working in two coordinated multi-robot tasks. In these, we contrasted standard displays, which assume each robot is independent, with an ecological socially-attentive display that makes the state of coordination explicit. The results show significant improvements in task completion time, number of failures, and the failure rate. Moreover, the display reduces the variance in operator control, thus leading to significantly more consistent operator performance

Robots in formations move while maintaining a predefined geometric shape. Previous work has examined formation-maintenance algorithms that would ensure the stability of the formation. However, for each geometric formation, an exponential number of stable controllers exists. Thus a key question is how to select (construct) a formation controller that optimizes desired properties, such as sensor usage for robustness. This paper presents a monitoring multi-graph framework for formation controller selection, based on sensor-morphology considerations. We instantiate the framework, and present two contributions. First, we show that graph-theoretic techniques can then be used to compute sensing policies that maintain a given formation. In particular, sensor-based control laws for separation-bearing (distance-angle) formation control can be automatically constructed. Second, we present a protocol allowing controllers to be switched on-line, to allow robots to adjust to sensory failures. We report on results from comprehensive experiments with physical robots. The results show that the use of the dynamic protocol allows formations of physical robots to move significantly faster and with greater precision, while reducing the number of formation failures

Open distributed multi-agent systems are gaining interest in the academic community and in industry. In such open settings, agents are often coordinated using standardized agent conversation protocols. The representation of such protocols (for analysis, validation, monitoring, etc) is an important aspect of multi-agent applications. Recently, Petri nets have been shown to be an interesting approach to such representation, and radically different approaches using Petri nets have been proposed. However, their relative strengths and weaknesses have not been examined. Moreover, their scalability and suitability for different tasks have not been addressed. This paper addresses both these challenges. First, we analyze existing Petri net representations in terms of their scalability and appropriateness for overhearing, an important task in monitoring open multi-agent systems. Then, building on the insights gained, we introduce a novel representation using Colored Petri nets that explicitly represent legal joint conversation states and messages. This representation approach offers significant improvements in scalability and is particularly suitable for overhearing. Furthermore, we show that this new representation offers a comprehensive coverage of all conversation features of FIPA conversation standards. We also present a procedure for transforming AUML conversation protocol diagrams (a standard human-readable representation), to our Colored Petri net representation.

Area coverage is an important task for mobile robots, with many real-world applications. Motivated by potential efficiency and robustness improvements, there is growing interest in the use of multiple robots in coverage. Previous investigations of multi-robot coverage focuses on completeness and eliminating redundancy, but does not formally address robustness, nor examine the impact of the initial positions of robots on the coverage time. Indeed, a common assumption is that non-redundancy leads to improved coverage time. We address robustness and efficiency in a family of multi-robot coverage algorithms, based on spanning-tree coverage of approximate cell decomposition. We analytically show that the algorithms are robust, in that as long as a single robot is able to move, the coverage will be completed. We also show that non-redundant (non-back tracking) versions of the algorithms have a worst-case coverage time virtually identical to that of a single robot—thus no performance gain is guaranteed in non-redundant coverage. Moreover, this worst-case is in fact common in real-world applications. Surprisingly, however, redundant coverage algorithms lead to guaranteed performance which halves the coverage time even in the worst case.

Recent years are seeing an increasing need for on-line monitoring of teams of cooperating agents, e.g., for visualization, or performance tracking. However, in monitoring deployed teams, we often cannot rely on the agents to always communicate their state to the monitoring system. This paper presents a non-intrusive approach to monitoring by 'overhearing', where the monitored team's state is inferred (via plan-recognition) from team-members' *routine* communications, exchanged as part of their coordinated task execution, and observed (overheard) by the monitoring system. Key challenges in this approach include the demanding run-time requirements of monitoring, the scarceness of observations (increasing monitoring uncertainty), and the need to scale-up monitoring to address potentially large teams. To address these, we present a set of complementary novel techniques, exploiting knowledge of the social structures and procedures in the monitored team: (i) an efficient probabilistic plan-recognition algorithm, well-suited for processing communications as observations; (ii) an approach to exploiting knowledge of the team's social behavior to predict future observations during execution (reducing monitoring uncertainty); and (iii) monitoring algorithms that trade expressivity for scalability, representing only certain useful monitoring hypotheses, but allowing for any number of agents and their different activities to be represented in a single coherent entity. We present an empirical evaluation of these techniques, in combination and apart, in monitoring a deployed team of agents, running on machines physically distributed across the country, and engaged in complex, dynamic task execution. We also compare the performance of these techniques to human expert and novice monitors, and show that the techniques presented are capable of monitoring at human-expert levels, despite the difficulty of the task.

Agents in dynamic multi-agent environments must monitor their peers to execute individual and group plans. A key open question is how much monitoring of other agents' states is required to be effective: The Monitoring Selectivity Problem. We investigate this question in the context of detecting failures in teams of cooperating agents, via Socially-Attentive Monitoring, which focuses on monitoring for failures in the social relationships between the agents. We empirically and analytically explore a family of socially-attentive teamwork monitoring algorithms in two dynamic, complex, multi-agent domains, under varying conditions of task distribution and uncertainty. We show that a centralized scheme using a complex algorithm trades correctness for completeness and requires monitoring all teammates. In contrast, a simple distributed teamwork monitoring algorithm results in correct and complete detection of teamwork failures, despite relying on limited, uncertain knowledge, and monitoring only key agents in a team. In addition, we report on the design of a socially-attentive monitoring system and demonstrate its generality in monitoring several coordination relationships, diagnosing detected failures, and both on-line and off-line applications.