Reproduce “AutoWorld: Learning Multi-Agent Traffic Simulation with Self-Supervised World Models” 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
Mozhgan Pourkeshavarz, Tianran Liu, Nicholas Rhinehart
AutoWorld predicts scene occupancy from unlabeled LiDAR, then uses those forecasts to guide a model that generates multi-agent behavior. Occupancy captures changing regions without semantic labels or agent identities, while the motion model uses labeled trajectories. In nuPlan’s high-occlusion test, its collision rate is 14.63%, about 36 percentage points below Nexus’s 50.46%, a simulation result rather than evidence of real-world safety.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.