Using a contingency planner within a robot architecture

From an automated planning perspective the problem of practical mobile robot control in realistic environments poses many important and contrary challenges. On the one hand, the planning process must be lightweight, robust, and timely. Over the lifetime of the robot it must always respond quickly with new plans that accommodate exogenous events, changing objectives, and the underlying unpredictability of the environment. On the other hand, in order to promote efficient behaviours the planning process must perform computationally expensive reasoning about contingencies and possible revisions of subjective beliefs according to quantitatively modelled uncertainty in acting and sensing. Towards addressing these challenges, we develop a continual planning approach that switches between using a fast satisficing "classical" planner, to decide on the overall strategy, and decision-theoretic planning to solve small abstract subproblems where deeper consideration of the sensing model is both practical, and can significantly impact overall performance. We evaluate our approach in large problems from a realistic robot exploration domain.

Reasoning with Uncertainty in Robotics, 2005

In this paper we report on the integration of a high-level plan executor with a behavior-based architec- ture. The executor is designed to execute plans that solve problems in partially observable domains. We discuss the different modules of the overall architecture and how we made the different modules interact using a shared representation. We also give a detailed description of

From an automated planning perspective the problem of practical mobile robot control in realistic environments poses many important and contrary challenges. On the one hand, the planning process must be lightweight, robust, and timely. Over the lifetime of the robot it must always respond quickly with new plans that accommodate exogenous events, changing objectives, and the underlying unpredictability of the environment. On the other hand, in order to promote efficient behaviours the planning process must perform computationally expensive reasoning about contingencies and possible revisions of subjective beliefs according to quantitatively modelled uncertainty in acting and sensing. Towards addressing these challenges, we develop a continual planning approach that switches between using a fast satisficing "classical" planner, to decide on the overall strategy, and decision-theoretic planning to solve small abstract subproblems where deeper consideration of the sensing model is both practical, and can significantly impact overall performance. We evaluate our approach in large problems from a realistic robot exploration domain.

Automated planners have been employed in a variety of robotics applications. However, there still remains a distinct divide between task planning, or high-level planning, and its counterparts in robotics. In particular, navigation and dialogue planning have emerged as important concerns in the quest to make realistic end-to-end robotic systems a reality. However, in the absence of a unifying problem to solve, collaborations between these three fields have been sparse and mostly narrow and project-driven. In this paper, we discuss Human-Robot Teaming as that unifying problem, and outline via a simple example the various sub-fields of the different kinds of planning that interact naturally to produce a solution to the overall problem. Our hope is to spur the various, fragmented planning communities into further collaboration by highlighting the rich potential of these interactions.

1994

Agents situated in dynamic and unpredictable environments require several capabilities, including synthesizing and executing plans while continuing to be responsive to the world. The Cypress system is a domain-independent framework for defining persistent agents with this full range of behavior. Cypress has been used for several applications, including military operations and fault diagnosis on the Space Shuttle.

This paper proposes a fast alternative to POMDP planning for domains with deterministic state-changing actions but probabilistic observation-making actions and partial observability of the initial state. This class of planning problems, which we call quasi-deterministic problems, includes important realworld domains such as planning for Mars rovers.The approach we take is to reformulate the quasi-deterministic problem into a completely observable problem and build a contingency plan where branches in the plan occur wherever an observational action is used to determine the value of some state variable. The plan for the completely observable problem is constructed assuming that state variables can be determined exactly at execution time using these observational actions. Since this is often not the case due to imperfect sensing of the world, we then use execution monitoring to select additional actions at execution time to determine the value of the state variable sufficiently accurately. We use a value of information calculation to determine which information gathering actions to perform and when to stop gathering information and continue execution of the branching plan. We show empirically that while the plans found are not optimal, they can be generated much faster, and are of better quality than other approaches.

SRI has built a domain-independent planning and execution system called CYPRESS which satisfies the following requirements: imperfect input information, rapid response during execution, minimal allocated computing resources during execution, incorporate preplanned plans, integrate planning and execution, and special technique for optimization of schedules. CYPRESS contains SIPE-2, a classical hierarchial Al planning system; PRS-CL, a reactive plan execution system; Gister-CL, a suite of evidential reasoning tools and the Grasper-CL system for interface and graphics. CYPRESS also runs in the ARPA/Rome Laboratory Planning Initiatives Common Prototyping Environment (CPE) and uses that system for communication between subsystems

Robotics and Autonomous Systems, 2003

The goal of robotics research is to design a robot to fulfill a variety of tasks in the real world. Inherent in the real world is a high degree of uncertainty about the robot's behavior and about the world. We introduce a robot task architecture, DTRC, that generates plans with actions that incorporate costs and uncertain effects, and states that yield rewards.

2013 16th International Conference on Advanced Robotics (ICAR), 2013

The standardization of planning problems by their descriptions in the PDDL has resulted in clear benchmarking of planners, and thus, in significant advances in reliable and efficient planning packages. The output of these classical planners is a plan as sequence of actions for the controllable robots in the environment. We show here that, provided that the adversaries follow a deterministic behavior, PDDLplanners can also be used in dynamic environments where uncontrollable adversaries may obstruct paths at some time in the future. Therefore, these environments can be used by mobile robots without the need to use more sophisticated planners where environments are modeled by Markov Decision Processes (MDPs). We created a planning API for integrating any PDDLsolver and use it to elaborate platform independent planning behavior. We also have the ability of switching between PDDLsolvers or to change the integration cycle of the planner. We show that these two features are essential for the dynamic environments considered here.

International Joint Conference on Artificial Intelligence, 1995

With increased processor speed and improved robotic and AI technology, researchers are be ginning to design programs that can behave in telligently and interact in the real world. A large increase in processing power has come from parallel machines, but taking advantage of this power is challenging. In this paper we address the issues in designing planners for real-time AI and robotic applications, and provide guiding principles. These principles were designed to minimize the difference be tween the new real-time model and the stan dard off-line model. Applying these princi ples yields a better-structured application, eas ier design and implementation, and improved performance. The focus of the paper is on a design methodology for implementing effec tive planners in real-world applications. Using Ephor (our runtime environment), and apply ing the described planner principles, we demon strate improved performance in a real-world shepherding application.