NTU's World-Model-Centric Racing Agent Achieves 88.3% Interaction Success
jiqizhixin · x · 2026-08-25
Nanyang Technological University's AutoMan Lab introduces a world-model-centric autonomous racing agent. Instead of chaining perception, prediction, planning, and control, a world model predicts opponent interactions, ego dynamics, and feasible motion envelope. The agent imagines multiple future scenarios before acting, filtering risky moves. Real racing data, including near-crash failures, continuously retrains the model. Simulation results show 88.3% interaction success rate (40.8% above BC, 29.3% above SAC), with over 6.4 km average safe driving distance.
More from Embodied
- Humanoid robot runs 100m in 9.39s, nearing Bolt's record — heyshrutimishra · 2026-08-25
- Apple unveils new Mac Studio with M5 Ultra, up to 512GB unified memory — themixtergames · 2026-08-25
- Robotics Outlook: Intelligence Won't Be the Bottleneck Next Year, Supply Will — Rewkang · 2026-08-25
- Matic robot fails: Bread tie stuck in vacuum duct — tristanbob · 2026-08-25
- Humanoid robot invents new running style to avoid shoulder overheating — CSProfKGD · 2026-08-25
- China unveils voice-summoned self-driving robot toilet — Polymarket · 2026-08-25