Autonomous Driving Simulation
Overview
The field builds virtual driving environments for training autonomous vehicles, testing safety, and studying interactions among cars, pedestrians, and other road users. Approaches range from physics-based simulators and CARLA–SUMO co-simulation to learned 3D scene reconstructions, occupancy models, and action-conditioned video or sensor world models; scenario generation often uses search, diffusion models, reinforcement learning, or language models. Evaluation spans closed-loop benchmarks such as CARLA, Bench2Drive, nuPlan, and Waymo, as well as hardware-in-the-loop and mixed-reality tests. A central tension is making simulations realistic and physically consistent while keeping them controllable and efficient, and ensuring that success in simulation predicts safe performance under new conditions and in the real world.
Latest papers
230 papers15 papers in the last four weeks, down 50% on the four weeks before. 0.1% of all new papers.