An Open-Source Modular Benchmark for Diffusion-Based Motion Planning in Closed-Loop Autonomous Driving

Researchers developed an open-source modular benchmark for evaluating diffusion-based motion planners in real-world autonomous driving systems. The system integrates with Autoware ROS 2 stack and achieves 3.2x latency reduction through encoder caching while improving accuracy by 41% with second-order solving.

• →New open-source benchmark addresses gaps in evaluating diffusion-based motion planners within production autonomous driving stacks.
• →System decomposes monolithic 18,398 node diffusion planner into three independently executable modules using ONNX GraphSurgeon.
• →Integration with Autoware ROS 2 enables runtime-configurable parameters without model recompilation.
• →Encoder caching achieves 3.2x latency reduction in motion planning performance.
• →Second-order DPM-Solver++ reduces Final Displacement Error by 41% compared to first-order at N=3 steps.