Autonomous Driving Planning

Autonomous driving planning turns sensor and map information, navigation goals, and predictions about other road users into decisions and vehicle trajectories. Approaches range from search and model predictive control (MPC), reachability and barrier methods, and game-theoretic planning to imitation learning, reinforcement learning, diffusion planners, and vision-language-action or world-action models. Systems increasingly combine prediction with planning to handle interactions, uncertainty, occlusion, and rare hazards, and are evaluated on benchmarks such as nuPlan, nuScenes, NAVSIM, Bench2Drive, and Waymo. A central challenge is that good open-loop trajectory scores do not guarantee safe, efficient closed-loop driving, while stronger safety checks and richer models must still meet real-time constraints and generalize to unfamiliar scenarios.