Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

robo-harness

SO-101 robot-arm workbench where a Jev decision runner picks bounded joint steps from typed candidates under a spend cap.

Projectprototype
View on GitHub ↗

Overview

Robo Harness is a standalone SO-101 robot-arm playground with a custom agent loop, Rerun as the visual workspace, explicit robot tools and a LeRobot hardware boundary. It defaults to mock hardware with synthetic cameras; a lab deployment drives a real arm with two Pi cameras. You can jog joints or Cartesian XYZ manually, teleoperate with a leader arm, record observations and actions, and supervise chat-driven moves that cannot take over human control.

Decision runs (bun run jev) let an evaluation model or a rules baseline pick bounded joint steps from typed candidate actions, with strategies named choice, parallel, critic and rules, plus smoke, observe, fixtures and dry-run modes. Jev is called as an evaluation model through TanStack decide() on the Vercel AI Gateway, cumulative spend is capped by ROBO_JEV_BUDGET_USD (default $10), and executing on the real arm requires a supervised flag with an operator present. Chat and a decision run never move the arm at the same time.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich bounded joint step from the candidate set should the arm take next?—

How it works

  1. Observe arm state and generate bounded candidate steps
  2. Jev (or the rules baseline) chooses a step via a strategy such as choice or critic
  3. Execute under supervision with step and time limits
  4. Record observations and actions for replay; stop at the spend cap

Strengths

  • Bounded candidates keep motion constrained
  • Spend cap and supervised mode
  • Rules baseline for comparison
  • Mock hardware for development

Caveats

  • Jev via Vercel AI Gateway needs paid credits
  • Real-arm runs need an operator present
  • No licence stated

Getting started

bun install && uv sync --extra dev && bun run build && bun run dev; then bun run jev --smoke --mock.

At a glance

Type
Project
Status
prototype
Language
TypeScript
Runtime
Bun/Effect app with Python drivers
Has code
yes

Use cases and integrations

How curated lists describe it

  • Robotics: SO-101 arm workbench where a Jev decision runner picks bounded joint steps from typed candidate actions under a spend budget.

    yibie/awesome-jev

  • SO-101 robot-arm workbench combining Bun/Effect and Python drivers, using Jev for joint action constraints.

    logicrw/awesome-jev-projects

Frequently asked questions

What does Jev decide?

Which bounded joint step to take next from typed candidate actions.

How is spending controlled?

ROBO_JEV_BUDGET_USD caps cumulative Jev spend, default $10.

Can I try it without a robot?

Yes, a fresh checkout defaults to mock hardware and synthetic cameras.

Related projects

  • jev-askable-arm

    Simulated Franka arm that follows plain-English goals by having Jev chain about 30 hardcoded motion primitives, one per step.

    Agent skillprototypePython
  • jev-drone

    Camera-only MuJoCo quadrotor where Jev makes 2.5 Hz tactical judgments while code keeps flight control and a safety veto.

    ProjectprototypePython
  • jev-reflex-autonomy-lab

    Multi-drone simulation where Jev makes fast reflex decisions per drone and an optional slower LLM may advise but never takes control.

    AppprototypeTypeScript
  • DriveJev

    Open 4B System I driving model that scores behaviours like cruise, stop, yield and brake from three camera frames each decision.

    ProjectresearchPython
  • JEV Control Your Roboarm

    Experimental robot loop where Jev chooses the next verified action from perception-generated options for a language goal.

    ProjectresearchPython
  • Jev for Physical AI

    Measured runs of Jev triaging incidents for a simulated 10,000-robot warehouse fleet, with cost and a ModernBERT comparison.

    ServiceresearchPython