Goto

Collaborating Authors

Efficient Planning under Uncertainty with Macro-actions

Journal of Artificial Intelligence Research

Deciding how to act in partially observable environments remains an active area of research. Identifying good sequences of decisions is particularly challenging when good control performance requires planning multiple steps into the future in domains with many states. Towards addressing this challenge, we present an online, forward-search algorithm called the Posterior Belief Distribution (PBD). PBD leverages a novel method for calculating the posterior distribution over beliefs that result after a sequence of actions is taken, given the set of observation sequences that could be received during this process. This method allows us to efficiently evaluate the expected reward of a sequence of primitive actions, which we refer to as macro-actions. We present a formal analysis of our approach, and examine its performance on two very large simulation experiments: scientific exploration and a target monitoring domain. We also demonstrate our algorithm being used to control a real robotic helicopter in a target monitoring experiment, which suggests that our approach has practical potential for planning in real-world, large partially observable domains where a multi-step lookahead is required to achieve good performance.


Efficient Planning under Uncertainty with Macro-actions

Journal of Artificial Intelligence Research

Deciding how to act in partially observable environments remains an active area of research. Identifying good sequences of decisions is particularly challenging when good control performance requires planning multiple steps into the future in domains with many states. Towards addressing this challenge, we present an online, forward-search algorithm called the Posterior Belief Distribution (PBD). PBD leverages a novel method for calculating the posterior distribution over beliefs that result after a sequence of actions is taken, given the set of observation sequences that could be received during this process. This method allows us to efficiently evaluate the expected reward of a sequence of primitive actions, which we refer to as macro-actions. We present a formal analysis of our approach, and examine its performance on two very large simulation experiments: scientific exploration and a target monitoring domain. We also demonstrate our algorithm being used to control a real robotic helicopter in a target monitoring experiment, which suggests that our approach has practical potential for planning in real-world, large partially observable domains where a multi-step lookahead is required to achieve good performance.


Efficient Planning under Uncertainty with Macro-actions

Journal of Artificial Intelligence Research

Deciding how to act in partially observable environments remains an active area of research. Identifying good sequences of decisions is particularly challenging when good control performance requires planning multiple steps into the future in domains with many states. Towards addressing this challenge, we present an online, forward-search algorithm called the Posterior Belief Distribution (PBD). PBD leverages a novel method for calculating the posterior distribution over beliefs that result after a sequence of actions is taken, given the set of observation sequences that could be received during this process. This method allows us to efficiently evaluate the expected reward of a sequence of primitive actions, which we refer to as macro-actions. We present a formal analysis of our approach, and examine its performance on two very large simulation experiments: scientific exploration and a target monitoring domain. We also demonstrate our algorithm being used to control a real robotic helicopter in a target monitoring experiment, which suggests that our approach has practical potential for planning in real-world, large partially observable domains where a multi-step lookahead is required to achieve good performance.


Efficient Planning under Uncertainty with Macro-actions

Journal of Artificial Intelligence Research

Deciding how to act in partially observable environments remains an active area of research. Identifying good sequences of decisions is particularly challenging when good control performance requires planning multiple steps into the future in domains with many states. Towards addressing this challenge, we present an online, forward-search algorithm called the Posterior Belief Distribution (PBD). PBD leverages a novel method for calculating the posterior distribution over beliefs that result after a sequence of actions is taken, given the set of observation sequences that could be received during this process. This method allows us to efficiently evaluate the expected reward of a sequence of primitive actions, which we refer to as macro-actions. We present a formal analysis of our approach, and examine its performance on two very large simulation experiments: scientific exploration and a target monitoring domain. We also demonstrate our algorithm being used to control a real robotic helicopter in a target monitoring experiment, which suggests that our approach has practical potential for planning in real-world, large partially observable domains where a multi-step lookahead is required to achieve good performance.


An AI Planning-Based Approach to the Multi-Agent Plan Recognition Problem (Preliminary Report)

AAAI Conferences

Plan Recognition is the problem of inferring the goals and plans of an agent given a set of observations. In Multi-Agent Plan Recognition (MAPR) the task is extended to inferring the goals and plans of multiple agents. Previous MAPR approaches have largely focused on recognizing team structures and behaviors, given perfect and complete observations of the actions of individual agents. However, in many real-world applications of MAPR, observations are unreliable or missing; they are often over properties of the world rather than actions; and the observations that are made may not be explainable by the agents' goals and plans. Moreover, the actions of the agents could be durative or concurrent. In this paper, we address the problem of MAPR with temporal actions and with observations that can be unreliable, missing or unexplainable. To this end, we propose a multi-step compilation technique that enables the use of AI planning for the computation of the posterior probabilities of the possible goals. In addition, we propose a set of novel benchmarks that enable a standard evaluation of solutions that address the MAPR problem with temporal actions and such observations. We present results of an experimental evaluation on this set of benchmarks, using several temporal and diverse planners.