Reproduce “DriveMLP: When Training Strategy Outperforms Model Complexity in Autonomous Driving” 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
Alessandro Canevaro, Julian Schmidt, Hang Yu, Ye Jin, Sajad Marvi, Georg Martius, Julian Jordan
DriveMLP trains an ego-driving planner with Expert-Guided RL (EGRL), deriving dense rewards from logged expert trajectories while other agents follow recorded paths. On nuPlan Val14 Reactive, it scores 86.51, versus 78.05 for BC-RL with the same architecture. Training uses non-reactive log replay, so the results do not establish performance in fully reactive training or real-world driving.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.