// PROJECTS · 29 WEEKS

Ten projects from a home lab.

Start on the laptop, then move to the arm on the desk, then to a rover in the garage. Projects 1–7 are the spine. Projects 8–10 are optional depth if the core is solid. Each one ends with a folder you can open, a short clip, and a row in the shared eval log.

PLANNEDP09 · Advanced · advanced

Legged sim-to-real

Train locomotion at scale, ablate domain randomization on hardware, and attribute the gap to actuator limits you already measured.

Weeks 23–29 · $0 on Bittle · or +$1,600 Go2 · 09_legged_rl/

Isaac Lab (train) + MuJoCo (contact studies)Train · RTX 5090Run · ESP32 on Bittle, or Go2 onboardBuy · $0 if Bittle from P8 · $1,600+ for Go2 EDUROS · No (optional later for nav)Gazebo · No
Legged sim-to-real
PLANNED
DEMO 09 · PLANNED
DEMO 09 · PLANNEDVideo lands here when this milestone ships.
Isaac Lab replay plus the 2×2 transfer table and GR05.
Quadruped on a path
Quadruped on a path
Uneven rock
Uneven rock
Rubble
Rubble
Isaac LabPPODomain randBittle / Go2Gap report
01

Goal

Isaac Lab, thousands of envs, PPO, velocity-tracking. Start from unitree_rl_gym. Treat domain randomization as a measured experiment: with/without, multiple seeds, same held-out perturbations. Then on hardware, did the randomized policy actually transfer better? Reward-term changes are separate harness runs. Deploy through common/deploy/ from P5/P8. A large gap on Bittle is the result, not the failure, if you can attribute it to P8 servo numbers. Buy a Go2 only if legged becomes the specialization.

02

Why it matters

Papers assume randomization helps. Measuring it on your own machine is the interesting part. “I trained a locomotion policy, deployed it on a $300 quadruped, and here is which actuator limit ate the transfer” is a better conversation than a clean Go2 clip that just reproduces someone’s repo.

03

The work

  1. 01

    Isaac Lab PPO velocity-tracking. Start from unitree_rl_gym.

  2. 02

    Domain-rand ablation: with/without × 3 seeds, identical held-out perturbations, confidence intervals.

  3. 03

    Every reward-term change is its own harness run with a before/after clip.

  4. 04

    Export through common/deploy/. Run on Bittle (or Go2).

  5. 05

    2×2 transfer: velocity-tracking error and fall rate, sim vs real, randomized vs not. GR05.

04

Done looks like

  • Trained locomotion policy with a headful Isaac Lab replay video.
  • Domain-rand ablation: with/without × 3 seeds, identical perturbations, confidence intervals.
  • Policy exported and running on real quadruped hardware.
  • Transfer 2×2: tracking error and fall rate, sim vs real, randomized vs not.
  • docs/gap_reports/GR05-legged-transfer.md attributing the gap to P8 actuator limits.
05

Stack card

SimulatorIsaac Lab (train) + MuJoCo (contact studies)
Trains onRTX 5090
Runs onESP32 on Bittle, or Go2 onboard
Buy$0 if Bittle from P8 · $1,600+ for Go2 EDU
ROS 2No (optional later for nav)
GazeboNo
Compute / targetESP32 (Bittle) or Go2 onboard
InstallIsaac Lab, rsl_rl, unitree_rl_gym as reference, ONNX export, Bittle or Unitree SDK
06

Repos to study

07

Watch out

  • Do not call a Bittle result a Go2 result.
  • ROS 2 only if you later add a navigation layer on top.
08

Workstreams

BUILD STREAM

Isaac Lab training, export, onboard bring-up.

EVAL STREAM

Ablation, 2×2 transfer, GR05.

Cross-check. Line-by-line review. This deploy path is reused in P10.

09

What this proves

You can train at scale, reason about what transfers and why, and produce evidence rather than a claim.

A large, attributed gap is a result. An unmeasured clean clip is a claim.

10

Study videos

11

Gear / BOM

Unitree Go2 EDU (optional)
Unitree Go2 EDU (optional)Buy only if P9 becomes the specialization. Default is Bittle.
$1,600+OPTIONAL
Gears: hardware + stack →