Skip to content

robot-control-stack: evaluating a VLA policy on real robots #335

Description

@Noietch

Hi maintainers,

I came across RobotControlStack/robot-control-stack and noticed that it focuses on “A lean, ROS-free sim-to-real framework for training and deploying Vision-Language-Action (VLA) models and RL agents. Native MuJoCo Gymnasium wrappers with synchronous execution for Franka, UR5e, xArm, SO101 and YAM.”. That is close to a problem we are working on with EVA-CLIENT, a unified framework for real-robot deployment, evaluation, and LeRobot data collection.

If this direction looks relevant, you can take a look at EVA-CLIENT here: https://github.com/Noietch/EVA-CLIENT

For RobotControlStack/robot-control-stack, EVA could provide the real-robot side of the sim-to-real loop, including checkpoint sweeps, latency logging, and synchronized replay. A small first experiment could be to wrap one checkpoint as a policy backend and run a small prompt/task sweep. We would be happy to prepare a minimal adapter or example around your current interfaces, rather than asking you to change the project architecture.

If this is useful, I would be glad to compare notes or help with a small integration. If it is not relevant, no worries—and we will not follow up.

Best,
EVA-CLIENT Team

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions