Body Schema Policy (BSP)

Body Schema Policy (BSP) is a visuomotor imitation-learning policy for whole-body locomotion and manipulation. Joint and link tokens follow the robot’s kinematic tree and query image features to denoise action chunks. The authors evaluated BSP on Unitree G1 in simulation and on hardware and on a simulated Fetch mobile manipulator.

View full model description

Papers & reports (1)

Body Schema Policy: Body as the Vision–Action Interface for Efficient Robot Loco-Manipulation

Paper available

arXiv: 2611.22713DOI: 10.48550/arxiv.2611.22713
Authors & affiliations (5)

Author order: Original order

  1. Lechen Zhang

    Mohamed bin Zayed University of Artificial Intelligence

  2. Chenyuan Chen

    Mohamed bin Zayed University of Artificial Intelligence

  3. Muxin Zhou

    Mohamed bin Zayed University of Artificial Intelligence

  4. Zewen He

    Mohamed bin Zayed University of Artificial Intelligence

  5. Yoshihiko Nakamura

    Mohamed bin Zayed University of Artificial Intelligence

First published
2026-11-17
Latest paper revision
v1 · 2026-11-17
Venue
Not provided

Covered model versions

Paper ↗PDF ↗

How to access and use

Training code for this version is unconfirmed. Public code resources do not establish access to the exact training implementation. View verification notes

Training / fine-tuning code

SourceVerified on · 2026-11-22

Release information

SourceVerified on · 2026-11-22

Checks before deployment

Training uses task-specific demonstrations. The project page lists the code as forthcoming. Public weights, the access route and the software and weights licenses remain unconfirmed. Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.

Deployment and usage resources

Inference mode
Closed-loop action chunking

Architecture and inputs / outputs

Model type
Action policy
Backbone
DINOv2 visual encoder
Action generation
Diffusion
Input
Language / RGB images / Robot state

Scope and training

Output

Actions

Prediction output

Denoised action chunks

Robot form

Humanoid / Mobile manipulator

Task scope

Multiple tasks

Training

Imitation learning

Related robots

Evaluation results

No verified independent benchmark results are recorded for this version or the selected weights.

Full parameters and verification notes (model card)

Unitree G1 · SIMPLE simulation (35.0M) 221/270 across nine tasks. 13.9 ms per 16-step chunk on RTX 6000 Ada.

Unitree G1 · physical robot (48.3M) 8/10 pushing a trolley and 8/10 using a pedal bin. 17.4 ms per chunk on RTX 5090.

Fetch · ManiSkill-HAB simulation (47.4M) SetTable: open fridge 34.4 ± 3.6%, close fridge 20.6 ± 1.6%, place 16.5 ± 1.6%. Mean ± SD over three seeds, 1,008 episodes per seed and subtask. The Fetch setup uses head and hand RGB cameras, joint state, and a language instruction.

The real G1 used a head-mounted RealSense D435i, two Dex3-1 hands and HoloMotion v1.4 to command 43 joints. Parameter counts are totals for each evaluated configuration.

Submit a source or correction