| The goal of robotics research is for robots to fulfill a variety of given tasks in the real world. Inherent in the real world is a high degree of uncertainty in the robot's behavior, and in information about the world. This work discusses a robot task architecture that generates action plans with actions that have costs and uncertainty, and states that grant rewards. This development leads to a more realistic application of planning suitable for robotic domains.; This thesis introduces DT-Graphplan, a decision-theoretic planner that generates plans for domains with incomplete information, stochastic action effects, state reward conditions, and action costs. The planner is used as one of three main elements in a robot task control architecture, DTRC (Decision-Theoretic Robot Controller) also discussed in this work.; The decision-theoretic planner, DT-Graphplan, is tested on several planning problems and the results are compared to two existing probabilistic planners and a POMDP (Partially Observable Markox Decision Problem) solver. Results from these experiments indicate that DT-Graphplan with pruning outperforms state-based decision theoretic approaches, such as Markov Decision Processes planners.; Our robot task control architecture contains three main elements: the DT-Graphplan planner, the robot skills, and the execution monitor. DT-Graphplan performs high level reasoning over the domain, generating action sequences to transition the agent from the initial condition to the goal condition. The robot skills are a collection of low-level base building blocks that handle motor and sensor control and integration, the highest level skills are the actions the planner uses to assemble a plan. The execution monitor communicates domain-information to DT-Graphplan and executes the plans generated by activating the robot skill for each plan step. Additionally, it maintains current domain information for the planner and detects plan failure and triggers replanning from the current condition.; The use of DT-Graphplan as a planner in a robot task architecture is demonstrated on the mobile robot domain of miniature golf. This shows the application of decision-theoretic planning in an inherently uncertain domain, and that by using decision-theoretic planning as the reasoning method in a task architecture, the architecture incorporates uncertain information in the reasoning process. |
