Sequential logic task planning method based on reinforcement learning

Abstract

The invention provides a sequential logic task planning method based on reinforcement learning. The sequential logic task planning method comprises the following steps: modeling a task by utilizing alinear sequential logic language and a non-deterministic Buchi automaton at first, describing an environment by utilizing a finite state transfer system FTS, and then generating a generative Buchi automaton by utilizing the FTS and the Buchi automaton; carrying out task planning training on the generative Buchi automaton by utilizing a Q-Learning method; and when the combined state enters the acceptable state or the dead zone state during the iterative training process, terminating the current round of iteration and jumping into the next iteration process, wherein a reward function utilized for updating the state-action value comprises the setting of the reward value or the penalty value given when the combined state enters the acceptable state or the dead zone state. By the adoption of the sequential logic task planning method provided by the invention, the effective, safe and high-speed task planning of the sequential logic task can be achieved.

Description

technical field [0001] The invention belongs to the field of artificial intelligence, and in particular relates to a temporal logic task planning method based on reinforcement learning. Background technique [0002] In actual tasks, many tasks not only require simple parallel cooperation, but also need to perform more complex serial cooperation tasks, that is, robots need to perform different tasks according to different stages of the task. Such tasks that require environment, time, and execution order are called sequential logic tasks. The task planning problem is to find a sequence of operations that change the system from an initial state to a goal state, given a task. The resolution process is similar to human deliberation, selecting and organizing actions by anticipating their outcomes. Mission planning can be applied in many fields: artificial intelligence, robotic systems, military command, etc. Therefore, the research in the field of mission planning attracts more...

AI Technical Summary

Problems solved by technology

[0006] The above scheme 1 needs to be calculated during use, and the real-time performance is not good; scheme 2 cannot handle new tasks that are not in the training data, and the effectiveness is insufficient, and the planning results are not safe, which may violate the given task timing logic

Images