Back to Home

Model-Based vs. Learning-Based Driving Simulation on WOMD/WOSAC

Built toolchain to auto-ingest WOMD map/scenarios into SUMO; long-horizon evaluation protocol; 0.653 realism on WOSAC’24.

Overview

Implemented automatic ingestion pipeline converting WOMD scenarios to SUMO, enabling system-level model-based multi-agent simulation for realism evaluation.

Trained SOTA simulation methods (SMART, TrafficBots) and showed complementary value with model-based SUMO (lower collision 0.47%, off-road 0.73%).

Media placeholder: Insert diagram of dataflow and sample SUMO/WOMD visualization.