armnet-busybox

The BusyBox environment for the Armnet managed arm farm.

BusyBox is an instrumented panel of switches, buttons, sliders and dials that reports its own state, so a cell built around one can tell whether a task succeeded without anyone watching, and put the panel back the way it started between episodes. This package is what turns that hardware into an Armnet environment: it reads the panel, decides whether an episode met its goal, and replays the taught motion that resets the scene.

Install

pip install armnet-busybox

You never import it. It registers itself under the armnet.environments entry point, and armnet-runtime finds it there when a job runs on a BusyBox cell. Installing it in your container is the whole of what you have to do:

from armnet_runtime import main

@main
def run(ctx):
    robot = ctx.cell.instrument(build_robot())   # panel readings attached
    ...
    if ctx.cell.is_complete():                   # scored by the panel
        ...
    ctx.cell.reset()                             # scene restored, no operator

Without it the same job still runs — armnet-runtime falls back to asking a human to judge and reset each episode.

Documentation

The environment, its tasks and its reference datasets: https://docs.armnet.dev/environments/busybox/

BusyBox itself is from Microsoft Research: paper and hardware.