This repository contains the code for the paper Obstacle-aware Waypoint Generation for Long-range Guidance of Deep-Reinforcement-Learning-based Navigation Approaches. It implements a waypoint generator, which considers dynamic obstacles for mid-range guidance of a DRL-based local planner. Therefore, Delaunay Triangles are utilized to encode dynamic obstacles and to extend a hybrid A* planner with a time domain. Subsequently the computed trajectory is optimized using the EGO Optimizer from Zhou et al. The approach is called Arena-FSM-EGO-Planner.
Arena-FSM-EGO-Planner is developed on top of the navigation framework arena-rosnav, which is a modular high-level library for end-to-end experiments in embodied AI -- defining embodied AI tasks (e.g. navigation, obstacle avoidance, behavior cloning), training agents (via imitation or reinforcement learning, or no learning at all using conventional approaches like DWA, TEB or MPC), and benchmarking their performance on the defined tasks using standard metrics. It uses Flatland as the core simulator.
![]() |
![]() |
---|---|
performance an a map with static obstacles | performance on an empty map |
Please refer to Installation.md for detailed explanations about the installation process.
Please refer to FSM-EGO-Planner for detailed explanations about the waypoint generators.