GigaBrain Challenge 2026 – Task 7 VLA Policy Model

This repository provides the Vision-Language-Action (VLA) policy model for Task 7 of the GigaBrain Challenge 2026.

Official Challenge Website:

https://gigaai-research.github.io/GigaBrain-Challenge-2026/

The GigaBrain Challenge aims to advance research in embodied intelligence, including:

  • Vision-Language-Action (VLA) models
  • world models
  • robot learning systems
  • closed-loop embodied agents

Task Overview

Task 7 is one of the official tasks in the GigaBrain Challenge 2026 benchmark suite.

Each task evaluates an embodied agent’s ability to generate actions based on multimodal observations and task instructions.

For full task descriptions please refer to:

https://gigaai-research.github.io/GigaBrain-Challenge-2026/index.html#tracks


Model Description

This repository contains the VLA policy model used for Task 7.

The model is designed to:

  • take visual observations and language instructions as input
  • generate robot actions
  • interact with a world model in a closed-loop evaluation environment

The VLA model serves as the action generation component in the challenge evaluation pipeline.

During evaluation: Observation → VLA Model → Action → World Model → Next Observation

This loop continues until the task is completed or the episode terminates.

Participants may use this model as:

  • a reference implementation
  • a starting point for model development
  • a component in their own embodied AI systems

Dataset

The dataset for all tasks is provided in the official challenge dataset repository:

https://huggingface.co/datasets/open-gigaai/CVPR-2026-WorldModel-Track-Dataset

Access requires:

  • Team registration
  • Agreement to the challenge license

Please request access on the dataset page.


License and Usage Terms

This model is released as part of the GigaBrain Challenge 2026.

Use of this model and the associated datasets is governed by the GigaBrain Challenge 2026 Data & Model License Agreement.

Key terms include:

Intended for research and competition participation

Dataset redistribution is not permitted

Commercial use requires permission from the organizers

For full license terms please refer to the dataset repository:

https://huggingface.co/datasets/open-gigaai/CVPR-2026-WorldModel-Track-Dataset

Citation

If you use this model or dataset in academic research, please cite the challenge:

@misc{gigabrain_challenge_2026,
  title={GigaBrain Challenge 2026},
  year={2026},
  howpublished={https://gigaai-research.github.io/GigaBrain-Challenge-2026/}
}

Disclaimer

This model is provided as-is without warranty of any kind.

The organizers are not responsible for any issues arising from its use.

Contact

For questions about the challenge, please visit the official website:

https://gigaai-research.github.io/GigaBrain-Challenge-2026/

Downloads last month
26
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Collection including open-gigaai/CVPR-2026-WorldModel-Track-Model-Task7