Reproduce “Autonomous Obstacle Removal for Excavators through Policy Learning with Particle Simulation” 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
Yuki Kadokawa, Sandro M. Alcantara Tacora, Taro Abe, Daisuke Endo, Genki Yamauchi, Takeshi Hashimoto, Takamitsu Matsubara
Researchers at Nara Institute of Science and Technology and Public Works Research Institute developed a policy learning framework for autonomous excavator obstacle removal using particle simulation. The method introduces a burial-conditioned curriculum to manage the computational cost of high-fidelity simulations, achieving over 90% success rates in real-world deployment and reducing training time by 50% compared to direct high-fidelity training.
Key insights
- Aligning task difficulty (obstacle burial condition) with simulation fidelity (particle count) significantly reduces the wall-clock training time for policies in high-fidelity particle simulations.
- A compact, parameterized action representation combined with dense visual-semantic observations allows policies to learn state-dependent, trajectory-level control for complex physical manipulation.
- Comprehensive domain randomization of soil, terrain, and obstacle properties is crucial for bridging the sim-to-real gap in tasks involving granular materials and intricate interactions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.