UniAD and UniAD 2.0

UniAD is a unified, planning-oriented autonomous driving framework that chains perception, prediction and planning tasks into one end-to-end model.

Made by OpenDriveLab, Hong Kong SAR China (checked on github.com/OpenDriveLab (opens in a new tab)).

Licenses

Code: Apache-2.0 (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit 609ee08 (opens in a new tab), on 16 Oct 2026.
Weights: Apache-2.0 (opens in a new tab)
Read on huggingface.co/OpenDriveLab/UniAD2.0_R101_nuScenes (opens in a new tab).
Commercial use: Yes

Casts perception, prediction and planning tasks hierarchically around a planning goal rather than as separate modules.
Trained in two stages: perception (track and map) first, then all task modules end-to-end.
Reports state-of-the-art results on motion prediction, occupancy and planning on the nuScenes benchmark.
UniAD 2.0 migrates the framework to mmdet3d 1.x and torch 2.x and adds nuPlan and NAVSIM dataset integration.
Pretrained checkpoints are offered for the UniAD-B model with an R101 encoder, with evaluation and training documented separately.
The NAVSIM planning results are given for an R34 encoder model, with a PDMS score of 83.4.

Written by Robotheca from the project's README (opens in a new tab) at commit 609ee08.

Checked on GitHub on 16 Oct 2026.

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.