Reproduce “Improving Policy Learning from Delayed Interventions” 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
Xiru Zhu, David Meger, Gregory Dudek
LfDI uses intervention timing, location, and coarse delay labels to infer risky earlier states, then combines cost-based policy learning with imitation of expert-controlled actions. In simulated Highway episodes, it traveled 927.7 ± 55.8 simulated units versus EIL’s 129.7 ± 9.0, with termination rates of 0.34 ± 0.06 and 1.00 ± 0.00. This result involved over 2,200 interventions; evaluations were simulated, not physical-robot deployments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.