The paper discusses a novel Deep Reinforcement Learning (DRL) algorithm for designing optimal control policies for robots in unknown environments. The algorithm uses a mission-driven exploration strategy to prioritize exploration directions that contribute to mission accomplishment. The Linear Temporal Logic (LTL) task is used to identify these directions, along with a neural network that models the unknown system dynamics. This approach results in faster learning rates compared to other methods. The efficiency of the algorithm is demonstrated through comparative experiments on robot navigation tasks.
Publication date: 29 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.17059
