hustvl/DiffusionDrive

Health v2 · maintenance only

Adoption not part of the score

1480 stars · 145 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

DiffusionDrive is a truncated diffusion model for real-time end-to-end autonomous driving, released as the official PyTorch implementation of a CVPR 2025 Highlight paper from HUST and Horizon Robotics. It achieves record-breaking 88.1 PDMS on NAVSIM with a ResNet-34 backbone while running at 45 FPS using only 2 denoising steps.

Use cases

• train an end-to-end driving policy on NAVSIM
• evaluate a diffusion-based planner on nuScenes
• generate diverse multimodal driving trajectories in real time
• reproduce CVPR 2025 autonomous driving benchmark results
• compare truncated diffusion policy against vanilla diffusion policy
• build a real-time motion planner for autonomous vehicles

When to choose

• you need a fast, multimodal end-to-end driving policy model
• you are benchmarking on NAVSIM or nuScenes
• you want to research diffusion-based planning for robotics or driving

When to avoid

• you need a production-ready autonomous driving stack
• you lack GPU resources for training diffusion models
• you need a general-purpose diffusion image generator

Facets

library · maturity active

machine-learning deep-learning computer-vision simulation autonomous-vehicles deep-learning computer-vision machine-learning python diffusion-model autonomous-driving end-to-end-driving navsim nuscenes cvpr-2025 motion-planning research-code gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="hustvl/DiffusionDrive")