Symbolic Sequence Optimization Approach for Task and Motion Planning of Robot Manipulators

In recent years, there has been a growing trend toward the use of robots to perform various tasks, such as setting out products in retail stores. Many new methods have been proposed for motion planning and object placement planning, however, there are few methods that optimize both of them simultaneously. In addition, if these methods are to be used in actual retail stores, users need to be able to easily give tasks to the robot. In this paper, we propose an optimization algorithm that allows a robot manipulator can optimize both object placement planning and motion planning at the same time from a simple input represented by a symbol sequence. We compare three types of search methods: random search, heuristic search with target state, combined heuristic and random search and UCT (Upper Confidence Tree). The results show that the combination of heuristic and random search is more efficient than other methods for searching state transitions, and is particularly effective for complex problems. In addition, comparing the conventional TAMP method with the proposed method, we found that the proposed method is able to search state transitions with a shorter robot operation time.