Reproduce “KlaskTron-IL: A Contact-Rich Adversarial Imitation Learning Benchmark with a Validated Sim-to-Real Bridge” 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
Nicklas Sohm, Aswin Ramachandran, Raffaello D'Andrea
KlaskTron-IL presents an imitation-learning benchmark for adversarial tabletop play, pairing a robot and simulator through a shared interface. The authors evaluate behavioral cloning, GAIL, and BC-initialized reinforcement learning using human and synthetic demonstrations. Increasing synthetic data from 80,000 to 5 million steps leaves BC about 0.015 below expert parity; zero-shot transfer is shown only for one BC policy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.