{"id":229,"date":"2021-07-12T14:33:27","date_gmt":"2021-07-12T14:33:27","guid":{"rendered":"https:\/\/fujie.ece.ufl.edu\/?page_id=229"},"modified":"2025-12-11T14:16:48","modified_gmt":"2025-12-11T19:16:48","slug":"learning-based-planning-with-temporal-logic-constraints","status":"publish","type":"page","link":"https:\/\/faculty.eng.ufl.edu\/fujie\/research\/learning-based-planning-with-temporal-logic-constraints\/","title":{"rendered":"Learning-based Planning with Temporal Logic Constraints"},"content":{"rendered":"

This\u00a0project is to develop a model-free reinforcement learning method for stochastic planning under temporal logic constraints. Using temporal logic formulas instead of reward function, we can rigorously express the desired properties to be achieved in the learned policies for stochastic systems. It will also provide provable performance guarantees in the controller learned from RL algorithms. With Temporal Logic, we can specify tasks such as, \u201cvisiting a sequence of waypoints A, B,C,D while avoiding obstacles\u201d, \u201cif the region A is visited, then region B must be visited before region C\u201d, \u201cif the probability of reaching A is less than 0.2, then only B,C,D need to be visited with a probability greater than 0.7.\u201d, and logical compositions of these properties.<\/p>\n