Installation#

Build pi and load the extension#

RoboAgentLab contains the full pi fork, its robotics extension, and Python services. Build the harness from this checkout so the CLI and extensions use the same source revision:

git clone https://github.com/RoboAgentLab/RoboAgentLab ~/RoboAgentLab
cd ~/RoboAgentLab
npm ci --ignore-scripts
node scripts/hydrate-ci-models.mjs
npm run build:offline
npm link --ignore-scripts --workspace @earendil-works/pi-coding-agent
pi install ./packages/embodied

The package initially loads only onboarding. /embodied-setup creates an experiment configured with one robot.

The catalog command downloads and verifies the pinned model catalog. build:offline then builds the repository’s packages and CLI from local sources. The first installation and catalog download require a network connection.

Check your local CLI#

command -v pi
ls -l "$(command -v pi)"
pi --version

The linked executable resolves to this checkout’s packages/coding-agent/dist/bundle/cli.js. If a different pi appears first on your PATH, put your npm global binary directory first or invoke the local build directly:

node ~/RoboAgentLab/packages/coding-agent/dist/bundle/cli.js

After pulling source updates, run npm ci --ignore-scripts, hydrate the pinned catalog, and rebuild. The global link continues to use the rebuilt CLI.

Install a backend manually#

From the repository root:

services/setup.sh libero --dry-run
services/setup.sh libero
source services/.venv-libero/pi-embodied.env
node packages/embodied/src/infra/check.ts libero --units

The setup script creates a dedicated virtual environment at services/.venv-<target>. The generated pi-embodied.env uses absolute paths, so it can be sourced before changing directories.

Option	Purpose
`--dry-run`	Print the installation plan.
`--weights`	Download the target’s model checkpoints.
`--no-assets`	Skip scene assets when already provisioned separately.
`--venv`	Choose the virtual environment location.
`--weights-dir`	Choose checkpoint storage.

See the service setup source for target-specific dependencies. Isaac Sim and Habitat backends also require their simulator installations and assets.

Use a remote GPU box#

In /embodied-setup, choose a remote GPU box over ssh, then give the SSH host and checkout location. Setup runs pi and the experiment on that host. You can also SSH in first and follow Quickstart there.

For services that are already running, use a deployment profile to select their endpoints and local Python environments.

Keep an experiment reproducible#

Record the platform commit, model identifier, backend, task, seed, mode, and budget alongside the result. Keep the source release when reproducing a fixed version:

git clone --branch v1.0 https://github.com/RoboAgentLab/RoboAgentLab RoboAgentLab-v1.0

The released tag stays fixed. Update a development checkout with git pull --ff-only, then rerun setup for a backend if its dependencies changed.

pi integration#

Robot tools, modes, and onboarding use pi’s extension interface. The locally built fork also supplies the session export and queued-message withdrawal hooks used by the dashboard. See the compatibility notes for the extension boundary.