Reproduce “Robot-Gated Interactive Imitation Learning with Adaptive Intervention Mechanism” 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
Haoyuan Cai, Zhenghao Peng, Bolei Zhou
Researchers at UCLA developed the Adaptive Intervention Mechanism (AIM), a robot-gated interactive imitation learning algorithm that learns when to request human assistance by mimicking human intervention patterns. AIM adaptively reduces expert involvement as the agent improves, achieving an 82% success rate in a driving simulator while requiring significantly fewer expert-involved steps compared to previous robot-gated methods.
Key insights
- Learning an intervention criterion based on a proxy Q-function can effectively mimic human intervention patterns, moving beyond simplistic uncertainty-based heuristics.
- Dynamically adjusting the intervention threshold based on the agent's improving performance (quantiles of Q-values) allows for adaptive and efficient human assistance without manual tuning.
- By focusing interventions on truly critical states, the approach collects higher quality expert data, leading to more robust policies with less human effort.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.