Poster Abstract: Signal Temporal Logic Compliant Motion Planning using Reinforcement Learning
Tushar Dilip Kurne, Manas Sashank Juvvi, J Vaishnavi, Pushpak Jagtap · 2024
This work proposes a novel approach for model-free motion planning in autonomous robots, with a specific emphasis on handling signal temporal logic (STL) tasks. The approach is structured into two main phases: first, we learn spatio-temporal motion primitives to encapsulate the inherent robot-specific constraints effectively. Subsequently, by utilizing these learned motion primitives, we conduct STL-compliant motion planning. To learn spatiotemporal motion primitives, we leverage reinforcement learning to construct a library of simple motion primitives. Following this, Gaussian process regression is employed to establish mappings from learned motion primitives to spatio-temporal characteristics. Subsequently, we present an STL-compliant motion planning strategy designed to meet the STL specification. Notably, the proposed framework is entirely model-free and capable of generating feasible STL-compliant motion plans across a diverse range of environments.