Skip to content
RobothecaBeta
Log in

Hy-Embodied-0.5 VLA

  • AI model
  • Humanoids and robot learning

An open vision-language-action model from Tencent that learns robot manipulation from large-scale egocentric demonstration data and transfers across robot platforms.

Made by Tencent Hunyuan.

Source on GitHub: github.com/Tencent-Hunyuan/Hy-Embodied-0.5-VLA (opens in a new tab). Facts read at commit af57e75 (opens in a new tab).

Licenses

Code
Apache-2.0 (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit af57e75 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Extends the Hy-Embodied-0.5 backbone with a flow-matching action expert that outputs continuous action chunks.
  • Predicts relative end-effector delta chunks, decoupling the policy from embodiment-specific kinematics.
  • A compact memory encoder compresses multi-frame, multi-view history into current-frame tokens without growing the token count.
  • Pre-trained on the Hy-UMI-10K corpus of UMI demonstrations; 2K+ hours of it are publicly released.
  • Released checkpoints are a UMI pre-trained generalist and a RoboTwin 2.0 post-trained model, with scripts for training and evaluation.
  • Includes an asynchronous inference stack and an adapter for the RoboDojo benchmark.

Written by Robotheca from the project's README (opens in a new tab) at commit af57e75.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
Linux (recommended), Python 3.12 (recommended and tested), CUDA 12.x, PyTorch 2.4 or newer, NVIDIA GPU with CUDA support (16 GB VRAM recommended); full pre-training is described for 64 GPUs.

Maintenance

Status
Active: a commit in the last 90 days
Newest commit
4 Aug 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots on Robotheca that use it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.