Amap · robot AI model
λ₀
China · 2026
Runs on G1 by Unitree RoboticsC
Sources retrieved: 27 October 2026 (A, B, C)
6 of 7 key figures stated, as stated by the developer
Specifications
9/14 specifications disclosed 5 not stated
2 fields hidden 3 not-stated fields with a note stay ·
Not statedThe developer publishes nothing on it
Select a source marker for source details and notes.
Release1/1 disclosed
Released13 October 2026
Source and note: ReleasedC arxiv.org/abs/2610.18950 · Retrieved 27 October 2026Note. [A] “Submitted on 13 Oct 2026”
Input and output3/4 disclosed
Model typeVision-language-action (VLA)
Source and note: Model typeA arxiv.org/html/2610.18950v1 · Retrieved 27 October 2026Note. Paper: “VLA training learns to predict actions from observations, language, and state using the resulting supervision.” The Hugging Face model tags include “vision-language-action”.
InputsCamera Language Robot state
Source and note: InputsA arxiv.org/html/2610.18950v1 · Retrieved 27 October 2026Note. Paper: “the policy takes an egocentric image, a language instruction, and the current state.”
What it controlsWhole body
Source and note: What it controlsA arxiv.org/html/2610.18950v1 · Retrieved 27 October 2026Note. Paper: “The predicted 64-dimensional motion token drives the SONIC whole-body controller, and the two 6-dimensional hand commands drive the Revo2 hands.”
Degrees of freedom controlledNot stated
Source and note: Degrees of freedom controlledA arxiv.org/html/2610.18950v1 · Retrieved 27 October 2026Note. Paper, Unitree G1 setup: “each step contains a 64-dimensional SONIC motion latent and two 6-dimensional Revo2 hand commands, totaling 76 dimensions”.
Size and speed0/2 disclosed
ParametersNot stated
Source and note: ParametersA arxiv.org/html/2610.18950v1 · Retrieved 27 October 2026Note. Paper: “The Qwen3.5-2B backbone supplies projected features to each layer of the action expert.”
Control rateNot stated
Where it runs and what it is built on1/2 disclosed
On robot or cloudNot stated
Base modelQwen3.5-2B
Source and note: Base modelB huggingface.co/lambda-humanoid/lambda0-midtrain-100k · Retrieved 27 October 2026Note. Model card: “The model builds on Qwen3.5-2B and learns from robot-aligned human demonstrations.”
Openness3/3 disclosed
AccessOpen weights (download)
Source and note: AccessB huggingface.co/lambda-humanoid/lambda0-midtrain-100k · Retrieved 27 October 2026Note. Hugging Face model card: “This is the Stage II mid-training checkpoint for λ₀.” “It does not contain EMA weights or a task-specific post-trained policy.” “This checkpoint is not a deployment-ready robot policy.” “The pretraining and mid-training datasets are not included.”
Licence documentApache-2.0
Source and note: Licence documentB huggingface.co/lambda-humanoid/lambda0-midtrain-100k · Retrieved 27 October 2026Note. Model card: “The λ₀ code and these weights are released under Apache 2.0.”
Licence typeFree, commercial use
Source and note: Licence typeB huggingface.co/lambda-humanoid/lambda0-midtrain-100k · Retrieved 27 October 2026Note. Model card: “The λ₀ code and these weights are released under Apache 2.0.”
The developer’s own claims1/2 disclosed
Learns new tasksNeeds retraining
Source and note: Learns new tasksB huggingface.co/lambda-humanoid/lambda0-midtrain-100k · Retrieved 27 October 2026Note. Developer claim: the model card sends adaptation to a post-training recipe: “Follow that recipe when it is released to adapt this checkpoint to your robot tasks.”
Works across robotsNot stated
Source and note: Works across robotsA arxiv.org/html/2610.18950v1 · Retrieved 27 October 2026Note. Paper: “equip our downstream robot, Unitree G1, with the same camera on its head”.Yes 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 λ₀ repository describes the model as learning whole-body humanoid loco-manipulation from everyday human experience.
The λ₀ project page lists the team as Alibaba Token Hub and Amap, Alibaba Group.
Company-register records (36Kr PitchHub, Maigoo) place Amap (AutoNavi) in Changping District, Beijing.
Sources
One line per source, not per figure: a source has one retrieval date.
- Aarxiv.org/html/2610.18950v1Retrieved 27 October 2026
- Bhuggingface.co/lambda-humanoid/lambda0-midtrain-100kRetrieved 27 October 2026
- Carxiv.org/abs/2610.18950Retrieved 27 October 2026