Peking University · robot AI model
Uni-NaVid
China · 2024
Sources retrieved: 18 November 2026 (A, B, C, D)
- Model type
- Vision-language-action (VLA)Source and note: Model typeD arxiv.org/abs/2412.06224 · Retrieved 18 November 2026Note. Paper abstract: Uni-NaVid is “the first video-based vision-language-action (VLA) model designed to unify diverse embodied navigation tasks”.
- Inputs
- Camera LanguageSource and note: InputsA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository: the “offline evaluation code of Uni-NaVid on real-world videos”; an example instruction is “(move to the chair, then turn left and move forward to the humanoid robot and stop.)”
- Access
- Open weights (download)Source and note: AccessC huggingface.co/api/models/Jzzhang/Uni-NaVid · Retrieved 18 November 2026Note. Hugging Face repository metadata list the file “uninavid-7b-full-224-video-fps-1-grid-2/config.json” and mark the repository as not gated. The code repository model table links the fine-tuned checkpoint “uninavid-7b-full-224-video-fps-1-grid-2”.
- Parameters
- 7BSource and note: ParametersA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository model table: “Uni-NaVid (7B)”; the pretrained language model it builds on is “Vicuna-7B”.
- Base model
- Vicuna-7BSource and note: Base modelA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository model table lists the pretrained language model as “Vicuna-7B”.

Specifications
8 of 14 stated 6 not stated
2 fields hidden 4 not-stated fields with a note stay ·
How to read the figuresShow guideClose
- not statedNo source
- 0Reported as zero
- ASource
Select a source marker for source details and notes.
Release1/1 disclosed
Released9 December 2024
Source and note: ReleasedD arxiv.org/abs/2412.06224 · Retrieved 18 November 2026Note. arXiv listing: “Submitted on 9 Dec 2024”.
Input and output2/4 disclosed
Model typeVision-language-action (VLA)
Source and note: Model typeD arxiv.org/abs/2412.06224 · Retrieved 18 November 2026Note. Paper abstract: Uni-NaVid is “the first video-based vision-language-action (VLA) model designed to unify diverse embodied navigation tasks”.
InputsCamera Language
Source and note: InputsA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository: the “offline evaluation code of Uni-NaVid on real-world videos”; an example instruction is “(move to the chair, then turn left and move forward to the humanoid robot and stop.)”
What it controlsNot stated
Source and note: What it controlsB pku-epic.github.io/Uni-NaVid/ · Retrieved 18 November 2026Note. The model outputs navigation actions, not motor commands: project page, the merged tokens and instruction tokens are sent to the large language model “to obtain actions for navigation or answers for embodied question-answering”.
Degrees of freedom controlledNot stated
Size and speed2/2 disclosed
Parameters7B
Source and note: ParametersA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository model table: “Uni-NaVid (7B)”; the pretrained language model it builds on is “Vicuna-7B”.
Control rateapproximately 5Hz
Source and note: Control rateA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Model inference rate, not motor control. Code repository: during evaluation the model is “achieving an inference speed of approximately 5 Hz on a single A100 GPU”.
Where it runs and what it is built on1/2 disclosed
On robot or cloudNot stated
Base modelVicuna-7B
Source and note: Base modelA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository model table lists the pretrained language model as “Vicuna-7B”.
Openness1/3 disclosed
AccessOpen weights (download)
Source and note: AccessC huggingface.co/api/models/Jzzhang/Uni-NaVid · Retrieved 18 November 2026Note. Hugging Face repository metadata list the file “uninavid-7b-full-224-video-fps-1-grid-2/config.json” and mark the repository as not gated. The code repository model table links the fine-tuned checkpoint “uninavid-7b-full-224-video-fps-1-grid-2”.
Licence documentNot stated
Source and note: Licence documentC huggingface.co/api/models/Jzzhang/Uni-NaVid · Retrieved 18 November 2026Note. The Hugging Face repository for the weights carries no licence. The code repository has a LICENSE file headed “MIT License”, which covers the code.
Licence typeNot stated
Source and note: Licence typeA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository: “This is an open-source version of Uni-NaVid”; the weights have no stated licence.
The developer’s own claims1/2 disclosed
Learns new tasksNeeds retraining
Source and note: Learns new tasksA github.com/jzhzhang/Uni-NaVid · Retrieved 18 November 2026Note. Code repository, training section: “If you want to finetune based on Uni-NaVid:”
Works across robotsNot stated
Source and note: Works across robotsYes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.
Notes
The project page lists the institutions as “1 Peking University”, “2 GalBot” and Beijing Academy of Artificial Intelligence; the page is hosted by the pku-epic group and names Peking University first.
Sources
One line per source, not per figure: a source has one retrieval date.
- Agithub.com/jzhzhang/Uni-NaVidRetrieved 18 November 2026
- Bpku-epic.github.io/Uni-NaVid/Main source. Retrieved 18 November 2026
- Chuggingface.co/api/models/Jzzhang/Uni-NaVidRetrieved 18 November 2026
- Darxiv.org/abs/2412.06224Retrieved 18 November 2026
More models
Same developer

LAWPeking University
- No photograph
LDA-1BPeking University
- No photograph
MM-NavPeking University
- No photograph
TrackVLAPeking University
- No photograph
TrackVLA++Peking University

UrbanVLAPeking University




