USVPlanner

Title: Local Collision Avoidance for Unmanned Surface Vehicles based on an End-to-End Planner with a LiDAR Beam Map Link

This paper has been published by IEEE Transactions on Intelligent Transportation Systems.

Introduction: This paper eliminates the need for cumbersome map maintenance and complex feature extraction by directly translating sensor data into navigational actions. A key innovation is the "beam map"—a novel observation modality that detects obstacles in all directions, mimicking onboard LiDAR. To address collision avoidance maneuvers in various encounter situations, a continuous-time short-distance constraint algorithm is designed to calculate COLREGs compliance rewards. This enables legal and rational navigation without requiring prior knowledge of the encounter scenario. Extensive experimental results, comparing various RL policies and classical methods, demonstrate the planner’s exceptional obstacle avoidance capability and adaptability to changing environments.

Beam map:

<img width="500" alt="beam-map" src="figure/beam-map.png" />

Running

1. Create a new environment

conda create -n usv_planner python=3.8

2. Install dependencies

pip install -r requirements.txt

3. Build and install RVO2

• Follow the steps from the official github

4. Start to train USV_planner with TD3 policy

python main_TD3.py

Device

• Ubuntu 20.04.6

<!-- ## Result --> <!-- ### 1. Trajectory comparison between different algorithms. --> <!-- Comparison algorithms:(a)DDPG (b)TD3 (c)DWA (d)APF --> <!-- <img width="500" alt="comparison" src="figure/comparison.png" /> --> <!-- ### 2. Generalization ability verification. --> <!-- <img width="700" alt="generalization" src="figure/generalization.png" /> -->

Acknowledgements

• The implementation of the beam map is inspired by the co-author zw199502's code.

• This project is based on AntoineTheb's implementation of the open-source solution of DRL structure.

• The ORCA algorithm implementation is from rebuttal-anonymous's code.

• The real inland AIS data is from FVessel.

References:

[1] W. Zhu and M. Hayashibe, "Learn to Navigate in Dynamic Environments with Normalized LiDAR Scans," 2024 IEEE International Conference on Robotics and Automation (ICRA), Yokohama, Japan, 2024, pp. 7568-7575, doi: 10.1109/ICRA57147.2024.10611247.