Plan

Marvin Bot is built in stages. Each stage ends with a pass check that decides whether to move on, and parts are bought only when the next stage needs them, so a wrong assumption costs one stage rather than the whole budget.

Principles

1. Simulation before hardware. The robot must stand, walk and climb stairs in simulation, with torques inside the actuators’ ratings, before more than two test actuators are bought.
2. Test the riskiest part with the smallest build: one actuator, then one leg, then the lower body, then the full robot.
3. Buy in stages. Spend is committed one phase at a time: the work column below says what each phase needs, and bom/bom.xlsx lists and prices every part (about $17.6k in all at November 2026 prices).
4. Measure, then update the model. Every hardware stage measures the real thing (torque, speed, heat, latency, mass) and feeds it back into description/ and the simulation.

Phases

#	Phase	Work	Pass check	Indicative time (part-time)
0	Foundations	Repo, requirements, decision log, licences, v0 design: joints, sizes, mass budget, BOM	Requirements and v0 layout agreed	1 week
1	Robot in simulation	Code-CAD of the frame, URDF and masses generated from it; standing, step-up and torque/thermal checks; walking and stair policies in Isaac Lab, cross-checked in MuJoCo	Walks and climbs 15–18 cm steps in simulation; peak torques ≤70% of rating; continuous torques within rating	3–5 weeks
2	Actuator bench	One RS04 and one RS03, CAN adapter, bench supply with regen clamp. Measure torque constant, speed, heating, CAN timing, built-in position control; fit actuator models and feed them back into simulation	Manual figures confirmed; models match measurements; CAN at the planned rates works on the Orin	2–3 weeks
3	One leg	Remaining actuators for one leg; machine and print one leg; test on a jig: trajectory tracking, ankle linkage range, gait-like loads for heat, simple hops	Leg tracks simulated motion; linkage reaches its range; nothing overheats	4–6 weeks
4	Lower body	Second leg, pelvis, battery and power system, e-stop, Orin, IMU. Stand on a gantry, walk on the gantry, then walk free	Walks at 0.5 m/s, turns, stops, recovers from pushes, handles low steps	6–8 weeks
5	Torso, arms, head	Waist, arms, wrist pitch and hands, neck, camera; whole-body controller; fall recovery	Walks while carrying objects; gets up after a fall	4–6 weeks
6	Perception and intelligence	Terrain map from the depth camera for stairs, teleoperation, data collection, GR00T N1.7 fine-tuning; finish the shells	Climbs real stairs using vision; picks up objects on command	ongoing

Workstreams

• Mechanical: code-CAD part generators, CNC aluminium cores, printed structural shells.
• Electrical: battery and BMS, power distribution, e-stop chain, CAN wiring.
• Onboard software: a real-time control process on the Orin (CAN I/O, policy inference at 50 Hz, safety) kept outside ROS; ROS 2 around it for the camera, logging and teleoperation.
• Learning: Isaac Lab tasks (velocity, rough terrain, stairs), MuJoCo cross-check, ONNX export; later motion tracking and the VLA.
• Docs: decision log, BOM, build guides, test results.

Repository growth

The current layout is in the README. Still to come, as phases reach them: docs/build/ (build guides, phase 3+), custom boards in electrical/ (phase 2+), the control loop and ROS 2 packages in robot/ (phase 2+), and summarised bench results in bench/results/ (phase 2). Trained policies are not committed; their training is documented in sim/isaaclab/README.md.

---

This page is generated from docs/plan.md in the repository.