Reproduce “Long-Range Indoor Navigation with PRM-RL” 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
Anthony Francis, Aleksandra Faust, Hao-Tien Lewis Chiang, Jasmine Hsu, J. Chase Kew, Marek Fiser, Tsang-Wei Edward Lee
Robotics at Google developed PRM-RL, a hierarchical navigation framework that integrates a learned reinforcement learning (RL) agent's robust local control capabilities directly into the construction of Probabilistic Roadmaps (PRMs). This approach enables robots to reliably navigate over long distances in complex, noisy indoor environments, demonstrating a low average sim2real gap of 7.43% on physical robots.
Key insights
- Embedding the probabilistic success rate of an RL-based local controller directly into PRM edge validity (via Monte Carlo rollouts) creates a kinodynamic planner that inherently accounts for robot dynamics, sensor noise, and local obstacle avoidance.
- Utilizing AutoRL to automate the search for optimal reward function parameters and neural network architectures significantly improves the quality and robustness of the learned local policies.
- Setting a higher success threshold for connecting PRM nodes (e.g., 100%) not only yields more reliable roadmaps but can also reduce construction time by enabling early termination of Monte Carlo trials for failed connections.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.