Skip to content
@ZGC-EmbodyAI

ZGC-EmbodyAI

ZGC-EmbodyAI

General Physical AI Project Group 🚀

Affiliated with Zhongguancun Academy (ZGCA) & Zhongguancun Institute of Artificial Intelligence (ZGCI)

Embodied AI VLA Models

📖 About Us

ZGC-EmbodyAI is a joint research group established by the Zhongguancun Academy (ZGCA) and the Zhongguancun Institute of Artificial Intelligence (ZGCI).

Our core mission is to develop General Physical AI.

🔬 Research Focus

  • 🤖 Post-training of LLMs & Agents: Advanced post-training techniques (e.g., SFT, RLHF) tailored for Large Language Models and autonomous agents to enhance reasoning and planning.
  • 🦾 VLA Architecture & Training: Designing and optimizing Vision-Language-Action (VLA) architectures for effective end-to-end robotic control.
  • 🌏 World Models in Embodied AI: Leveraging World Models to simulate, predict, and reason about physical interactions and dynamics.
  • 🛩︎ Real-world Reinforcement Learning: Deploying robust RL algorithms on physical hardware, including Robotic Arms, Dexterous Hands, and Humanoid Robots.

📝 Selected Publications

Below are our latest research contributions in the field of Embodied AI:

PhysBrain: Human Egocentric Data as a Bridge from Vision Language Models to Physical Intelligence

Abstract: An egocentric-aware embodied brain, PhysBrain, trained on E2E-3M, exhibits substantially improved egocentric understanding for planning tasks, provides sample-efficient VLA fine-tuning initialization, and achieves 53.9% SimplerEnv success rates—demonstrating effective transfer from human egocentric supervision to downstream robot control.

arXiv Code

TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformers

Abstract: TwinBrainVLA, a novel architecture that coordinates a generalist VLM retaining universal semantic understanding and a specialist VLM dedicated to embodied proprioception for joint robotic control.

arXiv Code

LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queries

Abstract: LangForce is a novel framework that enforces instruction following via Bayesian decomposition. By introducing learnable Latent Action Queries, LangForce construct a dual-branch architecture to estimate both a vision-only prior $p(a∣v)$ and a language-conditioned posterior $π(a∣v,\ell)$

arXiv Code


🤝 Affiliations

Beijing Zhongguancun Academy
(ZGCA)
Zhongguancun Institute of Artificial Intelligence
(ZGCI)

📍 Location: Beijing, China

📧 Contact: contact@example.com

Popular repositories Loading

  1. LangForce LangForce Public

    [ICML 2026] This repo is the official implementation of "LangForce : Bayesian Decomposition of Vision Language Action Models via Latent Action Queries"

    Python 81 5

  2. FrameSkip FrameSkip Public

    [EMNLP 2026 Findings] This repo is the official implementation of "FrameSkip: Learning from Fewer but More Informative Frames in VLA Training""

    Python 32 1

  3. TwinBrainVLA TwinBrainVLA Public

    30

  4. IntentVLA IntentVLA Public

    [EMNLP 2026] This repo is the official implementation of "IntentVLA: Short-Horizon Intent Modeling for Aliased Robot Manipulation"

    Python 14

  5. BayesianVLA BayesianVLA Public

    Forked from ZGC-EmbodyAI/LangForce

    Python 7

  6. PhysBrain PhysBrain Public

    HTML 6

Repositories

Showing 10 of 10 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…