Showing papers by "Brian C. Williams published in 2014"

Concurrent plan recognition and execution for human-robot teams

Abstract: There is a strong demand for robots to work in environments, such as aircraft manufacturing, where they share tasks with humans and must quickly adapt to each other's needs. To do so, a robot must both infer the intent of humans, and must adapt accordingly. The literature to date has made great progress on these two tasks - recognition and adaptation - but largely as separate research activities. In this paper, we present a unified approach to these two problems, in which recognition and adaptation occur concurrently and holistically. Key to our approach is a task representation that uses choice to represent alternative plans for both the human and robot, allowing a single set of algorithms to simultaneously achieve recognition and adaptation. To achieve such fluidity, a labeled propagation mechanism is used where decisions made by the human and robot during execution are propagated to relevant future open choices, as determined by causal link analysis, narrowing the possible options that the human would reasonably take (hence achieving intent recognition) as well as the possible actions the robot could consistently take (adaptation). This paper introduces Pike, an executive for human-robot teamwork that quickly adapts and infers intent based on the preconditions of actions in the plan, temporal constraints, unanticipated disturbances, and choices made previously (by either robot or human). We evaluate Pike's performance and demonstrate it on a household task in a human-robot team testbed.

Chance-constrained consistency for Probabilistic Temporal Plan Networks

Abstract: Unmanned deep-sea and planetary vehicles operate in highly uncertain environments. Autonomous agents often are not adopted in these domains due to the risk of mission failure, and loss of vehicles. Prior work on contingent plan execution addresses this issue by placing bounds on uncertain variables and by providing consistency guarantees for a 'worst-case' analysis, which tends to be too conservative for real-world applications. In this work, we unify features from trajectory optimization through risk-sensitive execution methods and high-level, contingent plan execution in order to extend existing guarantees of consistency for conditional plans to a chance-constrained setting. The result is a set of efficient algorithms for computing plan execution policies with explicit bounds on the risk of failure. To accomplish this, we introduce Probabilistic Temporal Plan Network (pTPN), which improve previous formulations, by incorporating probabilistic uncertainty and chance-constraints into the plan representation. We then introduce a novel method to the chance-constrained strong consistency problem, by leveraging a conflict-directed approach that searches for an execution policy that maximizes reward while meeting the risk constraint. Experimental results indicate that our approach for computing strongly consistent policies has an average scalability gain of about one order of magnitude, when compared to current methods based on chronological search.

A scheduler for actions with iterated durations

Abstract: A wide range of robotic missions contain actions that exhibit looping behavior. Examples of these actions include picking fruit in agriculture, pick-and-place tasks in manufacturing and search patterns in robotic search or survey missions. These looping actions often have a range of acceptable values for the number of loops and a preference function over them. For example, during robotic survey missions, the information gain is expected to increase with the number of loops in a search pattern. Since these looping actions also take time, which is typically bounded, there is a challenge of maximizing utility while respecting time constraints. In this paper, we introduce the Looping Temporal Problem with Preference (LTPP) as a simple parameterized extension of a simple temporal problem. In addition, we introduce a scheduling algorithm for LTPPs which leverages the structure of the problem to find the optimal solution efficiently. We show more than an order of magnitude improvement in run-time over current scheduling techniques and framing a LTPP as a MINLP.

A Demonstration of Robust Planning and Scheduling in the Techsat-21 Autonomous Sciencecraft Constellation

Abstract: The demonstration (ASC) will fly onboard the Air Force’s TechSat-21 constellation (an unclassified mission scheduled for launch in 2004). ASC will use onboard science analysis, replanning, robust execution, model-based estimation and control, and formation flying to radically increase science return by enabling intelligent downlink selection and autonomous retargeting. Demonstration of these capabilities in a flight environment will open up tremendous new opportunities in planetary science, space physics, and earth science that would be unreachable without this technology. We offer a demonstration of the planning, scheduling, and execution framework for this application.

General probabilistic bounds for trajectories using only mean and variance

Abstract: Two ideas have gained traction in research in the robotics planning community. Activity planning has become popular where a library of predefined manipulation of the vehicle state is accessible, and is commonly used for missions with complex goal specifications. Another focus has been chance-constrained programming as a method of providing robust motion planning, in which the probability of failure is bounded. A combination of the two would allow for robust satisfaction of complex directives. However, to perform chance-constrained activity planning, we must be able to provide probabilistic bounds on the trajectory of the vehicle. While this may be done through propagation of statistics, we would require information about the actuation noise for the vehicle dynamics. In addition to such parameters as mean and variance, we also need to know the appropriate function for the noise. In many cases, the exact distribution of the actuation noise may not be known, although researchers can easily approximate the first two moments through calibrations. In this work we look at statistics propagation when only the first two moments of the actuation uncertainty is known, assuming white noise. We show that for linear systems, propagation is exact. Further, by looking at the expected error squared as a stochastic process, we can show that it is a submartingale under certain assumptions, and thus derive error bounds for deviation from mean over the duration of the entire path. We empirically show that, for nonlinear dynamics, we may approximate the propagation with the unscented transform, and obtain the corresponding bounds.

Computational Sustainability: Editorial Introduction to the Summer and Fall Issues

Abstract: Computational sustainability problems, which exist in dynamic environments with high amounts of uncertainty, provide a variety of unique challenges to artificial intelligence research and the opportunity for significant impact upon our collective future. This editorial introduction provides an overview of artificial intelligence for computational sustainability, and introduces the special issue articles that appear in this issue and the previous issue of AI Magazine.

A new filter for hybrid systems and its applications to robust attitude estimation

Abstract: Fault diagnosis and recovery are essential tools for the development of autonomous agents that can operate in hazardous environments. This can be effectively approached from a model-based perspective, where sensor faults are explicitly taken into account in a hybrid model with switching dynamics. However, practical hybrid filters are required to manage an exponential growth in the number of discrete mode sequences, also known as hypotheses. Inspired by an attitude estimation application for a quadrotor UAV with faulty sensors, this paper introduces the IP-MHMF, a novel filter for hybrid systems that generalizes the well-known IMM and introduces a more informed hypothesis-pruning step than previous algorithms. By performing hypothesis pruning on corrected rather than predicted hypothesis probabilities, the IP-MHMF is capable of much more aggressive pruning strategies that significantly reduce its computational load, while improving its estimation performance. Our numerical results on data from a real robotic platform show that the IP-MHMF outperforms state-of-the-art hybrid filters and the traditional EKF on an attitude estimation application with faulty magnetometer measurements.

Drake: An Efficient Executive for Temporal Plans with Choice

Abstract: This work presents Drake, a dynamic executive for temporal plans with choice. Dynamic plan execution strategies allow an autonomous agent to react quickly to unfolding events, improving the robustness of the agent. Prior work developed methods for dynamically dispatching Simple Temporal Networks, and further research enriched the expressiveness of the plans executives could handle, including discrete choices, which are the focus of this work. However, in some approaches to date, these additional choices induce significant storage or latency requirements to make flexible execution possible. Drake is designed to leverage the low latency made possible by a preprocessing step called compilation, while avoiding high memory costs through a compact representation. We leverage the concepts of labels and environments, taken from prior work in Assumption-based Truth Maintenance Systems (ATMS), to concisely record the implications of the discrete choices, exploiting the structure of the plan to avoid redundant reasoning or storage. Our labeling and maintenance scheme, called the Labeled Value Set Maintenance System, is distinguished by its focus on properties fundamental to temporal problems, and, more generally, weighted graph algorithms. In particular, the maintenance system focuses on maintaining a minimal representation of non-dominated constraints. We benchmark Drakes performance on random structured problems, and find that Drake reduces the size of the compiled representation by a factor of over 500 for large problems, while incurring only a modest increase in run-time latency, compared to prior work in compiled executives for temporal plans with discrete choices.