Cooperative Motion Planning in Divided Environments via Congestion-Aware Deep Reinforcement Learning
This repository contains the official implementation of the paper:
"Cooperative Motion Planning in Divided Environments via Congestion-Aware Deep Reinforcement Learning"
Authors: Yuanyuan Du, Jianan Zhang, Xiang Cheng, and Shuguang Cui
Published in IEEE Robotics and Automation Letters (RA-L), December 2024.
This work proposes a novel cooperative motion planning algorithm leveraging Congestion-Aware Deep Reinforcement Learning (CCADRL) to address collisions and congestion in environments divided by narrow hallways. Key contributions include:
- A temporal arrival intent sharing paradigm that is used for constructing a hallway map, informing asynchronous individual motion planning around hallways.
- A non-myopic congestion-aware scheme that incorpo rates a hallway goal chooser and a congestion predictor. This scheme prevents the agent from adhering to heavily congested trajectories that may be only slightly shorter and enables the agent to decide whether to claim getting into or avoid the selected hallway.
- A relation analyzer that encodes interaction dynam ics among neighboring agents, enriching the agents’ decision-making capabilities.
Simulations demonstrate significant improvements over state-of-the-art algorithms in various challenging scenarios.
- Code Upload in Progress: The full implementation will be uploaded by the end of this week.
- This repository will include:
- Evaluation scripts.
- Simulated environments.
- Pre-trained models and detailed documentation.
Comprehensive instructions will be provided once the upload is complete. Check back soon for:
- Installation steps.
- Examples for testing CCADRL.
If you find this repository helpful, please consider citing our paper: