Skip to content
RobothecaBeta
Log in

VLAC

  • AI model
  • Humanoids and robot learning

VLAC is a pair-wise critic and manipulation model for real-world robot reinforcement learning and data refinement, scoring task progress and completion from images and a task description.

  • Non-commercial weights

Made by Intern Robotics, Shanghai AI Laboratory.

Source on GitHub: github.com/InternRobotics/VLAC (opens in a new tab). Facts read at commit 0f302d0 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 0f302d0 (opens in a new tab), on 16 Oct 2026.

Commercial use
No

The weights license does not allow commercial use.

Labels

  • Non-commercial weightsThe weights license (CC-BY-NC-SA-4.0) does not allow commercial use.

What it does

  • Compares pairs of frames to predict dense task progress and recognise state changes.
  • Verifies whether a task has been completed from images and a text description.
  • Supports visual question answering and can output embodied actions, with zero-shot and one-shot in-context use.
  • Screens collected trajectories, filtering low-scoring or low-fluency data to improve imitation learning.
  • A VLAC-cut variant cuts out bad or good segments from embodied trajectories and comes with a progress-annotated dataset.
  • Installed from source with pip; requires Python 3.9 or newer (3.10 recommended), PyTorch 2.0 or newer, and transformers 4.51 or newer.

Written by Robotheca from the project's README (opens in a new tab) at commit 0f302d0.

Getting started

Difficulty
Advanced

Maintenance

Status
Active: a commit in the last 90 days
Newest commit
18 Sept 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.