hustvl/VAD

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1362 stars · 169 forks observed · 2026-08-28

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

VAD is an end-to-end autonomous driving framework that models the driving scene as a fully vectorized representation of agents and map elements for trajectory planning. It includes VADv1 (ICCV 2023) and VADv2 (ICLR 2026), achieving state-of-the-art planning performance on nuScenes with faster inference than rasterized approaches.

Use cases

• run end-to-end trajectory planning on nuScenes
• reproduce vectorized autonomous driving research results
• train an end-to-end driving model with vectorized scene representation
• evaluate planning safety and collision rates for autonomous vehicles
• integrate VADv2 probabilistic planning into a driving stack
• benchmark inference speed of end-to-end driving planners

When to choose

• you need a fast, vectorized end-to-end planning baseline for autonomous driving research
• you want reproducible nuScenes planning benchmarks with pretrained models
• you are exploring VADv2-style probabilistic trajectory planning

When to avoid

• you need a production-ready, safety-certified driving stack
• your project requires rasterized occupancy-based scene representation
• you lack GPU resources for training large perception-planning models

Facets

library · maturity active

machine-learning deep-learning computer-vision simulation autonomous-vehicles machine-learning computer-vision robotics python end-to-end-driving vectorized-scene-representation trajectory-planning nuscenes iccv-2023 research-code vadv2 motion-planning gpu linux

Member repositories

For agents

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