An Approach to Agent Path Planning Under Temporal Logic Constraints

The capability of path planning is a necessity for an agent to accomplish tasks autonomously. Traditional path planning methods fail to complete tasks that are constrained by temporal properties, such as conditional reachability, safety, and liveness. Our work presents an integrated approach that combines reinforcement learning (RL) with multi-objective optimization to address path planning problems with the consideration of temporal logic constraints. The main contributions of this paper are as follows. (1) We propose an algorithm LCAP \(^2\) to design extra rewards and accelerate training by tackling a multi-objective optimization problem. The experimental results show that the method effectively accelerates the convergence of the path lengths traversed during the agent’s training. (2) We provide a convergence theorem based on the fixed-point theory and contraction mapping theorem.