Small decisions. Interesting possibilities.Submit contentSubmit

MakerMods robot arm in MuJoCo

Got @typesafeai Jev driving our @makermodsai Metal arm in MuJoCo. Images are of the sim camera is not passed in, sim hands Jev the state of the environment in JSON and it picks the next bounded action (hover, descend, grasp, lift, place) as a typed choice with a confidence score. Jev makes the decisions for example…

Added to Jevfast

How it uses Jev

The MuJoCo simulation sends Jev the environment state as JSON, without passing the simulation camera image. Jev selects a typed bounded action such as hover, descend, grasp, lift, or place. The simulator executes the action; the post says the example red-apple-to-blue-bin task took nine decisions and that Jev refuses tasks it cannot map to the scene.

App workflow

  1. Describe the simulated state

    MuJoCo supplies the environment state as JSON; the simulation camera image is not passed to Jev.

  2. Select a bounded action

    Jev chooses among actions such as hover, descend, grasp, lift, and place, or refuses an ungrounded task.

  3. Execute in simulation

    The simulator applies each choice; the cited apple-to-bin task completed in nine decisions.

Why it is interesting

It exposes a clear control loop from structured simulator state to bounded robot actions, with refusal when a task cannot be grounded in the scene.