Getting Started

Generated from docs/Getting-Started.md at 13d96a0e66b24bec83dc21a6f33149721973e483. Edit repository docs, not this mirror. This snapshot does not imply a merge to main.

Getting Started

Home | Contributing | Baseline audit

Work On The Foundation

Reading the Markdown needs only a browser or editor. Documentation checks use Git and Node.js 22 or newer. Node is documentation tooling, not a choice of desktop application language.

For a new checkout:

git clone https://github.com/emb-ai/FURRY.git
cd FURRY
git switch codex/20261005-project-foundation
npm ci --ignore-scripts
npm run docs:lint
npm run docs:check
npm test
git diff --check

The bootstrap documentation is on that branch while its PR awaits review; after merge, start new work from the repository's current default branch. In the owner's existing workspace, do not clone again. Inspect git status --short --branch and the current worktree before editing.

Choose an umbrella from the roadmap, then propose one bounded task, an expected artifact and a test. Read the evidence vocabulary before reporting results.

Prerequisites By Workstream

The table below is a planning guide, not a validated FURRY runtime setup.

Workstream	Likely tools or hardware	Gate before installation
Docs, source audit, experiment design	Git, Node 22+ for checks	None; macOS or Linux is sufficient for documentation.
Released TWIST2 reproduction	Linux/NVIDIA reference environment, upstream Python environments, MuJoCo, ONNX Runtime, Redis	Read the pinned audit, select a host, and record versions.
Quest acquisition	Quest 3, developer access, Android build/deploy tools; SDK/engine depends on reuse result	Check device capability and licensing; do not preselect an engine.
Controller-free hands	Optical-hand and body-tracking APIs where available	Record permissions, SDK/runtime versions, supported fields and invalidity signals.
RGB-D experiment	Available RealSense device and compatible driver/pose estimator	Inventory the actual camera/host before choosing packages.
Policy training	Separate GPU environment and licensed motion datasets	Only after baseline reproduction shows a concrete training need.

The upstream README splits Python 3.8 training/deployment and Python 3.10+ retargeting environments because of Isaac Gym dependencies. The released offline motion server itself imports Isaac Gym and constructs CUDA tensors. This prevents assuming that a CPU-capable low-level ONNX loop makes the entire pipeline portable. Pinned sources

macOS And Linux

The bootstrap machine is Apple Silicon macOS. Documentation tools are checked here; no robotics dependency stack is installed by this task. Upstream Linux instructions must not be pasted into macOS unchanged.

For Linux reproduction, use a separate environment and pin the upstream commit. Do not delete existing Conda environments to follow a README verbatim. Keep Redis on loopback for an offline baseline. Remote streaming requires a separate network-access decision, not disabling protected mode or the firewall.

A macOS/CPU runtime is an investigation, not a supported platform declaration. The low-level loop's CPU option is only one piece; a portable reference generator, viewer invocation and dependency set still need validation. MuJoCo's macOS passive viewer has specific launch requirements documented by MuJoCo.

Keep The Checkout Small

Store third-party checkouts beside FURRY, not inside it. Put private recordings, models, calibration captures and participant data outside the repository; reserved local data/, recordings/, checkpoints/ and third_party/ directories are ignored as an additional guard, not permission to redistribute their contents.

Do not commit headset serials, local network addresses, access tokens or private student identifiers. Obtain permission before collecting identifiable footage. Never run sim2real.sh, physical-robot SDK setup or robot network instructions as part of this project.

First Deliverable

Before a large implementation, submit a small source audit or experiment plan: one question, pinned sources, required inputs, success/failure criteria, and the next decision it enables. Use the experiment template.