Reproduce “COLREG-Compliant Collision Avoidance for Unmanned Surface Vehicle using Deep Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eivind Meyer, Amalie Heiberg, Adil Rasheed, Omer San
Researchers from NTNU and OSU developed a Deep Reinforcement Learning agent using Proximal Policy Optimization to enable Unmanned Surface Vehicles to simultaneously perform path following and COLREGs-compliant collision avoidance. The agent demonstrates adherence to international collision regulations and generalizes to complex real-world marine environments reconstructed with high-fidelity terrain and historical AIS data.
Key insights
- A single DRL agent can effectively integrate and balance the competing objectives of path following and COLREGs-compliant collision avoidance for an underactuated ASV.
- Innovative observation space design, including non-uniform sensor partitioning and feasibility pooling, is critical for enabling DRL agents to process complex environmental information efficiently.
- Complex regulatory rules like COLREGs can be successfully integrated into a DRL agent's behavior through careful reward function design and dynamic objective weighting, leading to realistic and compliant autonomous navigation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.