Treating the humanoid torso as a central compute and power backplane with a
wheeled MVP is a highly pragmatic development path. It completely decouples
the brutal physics of bipedal balancing (the inverted pendulum problem) from
the system architecture, allowing you to validate your software stack, CAN FD
topologies, and power distribution early.
By starting with a rover, you guarantee a functional, testable platform at
every milestone, preventing the project from stalling in the "simulation and
control theory" phase.
Here is how to architect this modular "transformer" progression:
The Universal Socket Interface
For the modularity to work, the "sockets" at the shoulders and hips must act
like hardware APIs. Instead of hardwiring limbs, you design a unified physical
and electrical bus.
 * Electrical Backplane: Each socket exposes high-current DC (24V or 48V), CAN
FD (High/Low), GND, and a hardware E-stop interrupt.
 * Actuation Standard: Use Quasi-Direct Drive (QDD) BLDC motors (e.g., AK80-9
style) directly at the sockets.
 * Mode Switching: In Stage 1 (Rover), the RP2040 motor controllers run in
velocity mode to spin the wheels. In Stage 2 (Humanoid), the exact same motors
switch to position/impedance control to act as the primary shoulder/hip pitch
joints.
The Flip-Proof Geometry
To allow the MVP to drive regardless of how it lands, the geometry requires
specific constraints:
 * Torso Depth vs. Wheel Diameter: The radius of the wheels must exceed the
depth (Z-axis) of the torso. If the torso houses a battery pack and compute
module and is 150mm thick, your wheels must be at least 160mm in diameter so
the chassis never bottoms out when inverted.
 * Symmetric Sensor Placement: If the rover is reversible, sensors (cameras,
IMUs, LIDAR) must either be omnidirectional, mounted on a central 360-degree
turret (the "head"), or duplicated on the front and back planes.
Realistic Milestone Pathway
Stage 1: The Core Node & Symmetric Rover
 * Hardware: Torso enclosure containing the battery, master controller, and
CAN FD hub. Four QDD motors bolted to the sockets with large, direct-drive RC
wheels attached.
 * Software Focus: Validating microsecond PTP time synchronization, ring
buffer telemetry, SLAM, and basic state machine routing. The system is
statically stable, so you can safely test edge-compute AI models and
navigation without breaking hardware.
Stage 2: The Articulated Rover (The "Centaur")
 * Hardware: Remove the wheels from the main sockets. Attach a 1-DOF limb (a
femur/humerus) with a second motor (knee/elbow) at the end. Reattach the
wheels to the ends of these limbs.
 * Software Focus: Introduces forward and inverse kinematics. The robot can
now change its ride height, lean into turns, and use active suspension. It
remains statically stable but forces you to master multi-joint CAN FD
coordination and spatial logic.
Stage 3: Bipedal Transition
 * Hardware: Swap the wheels on the lower limbs for feet (adding ankle
pitch/roll joints). Lock or swap the upper limbs for standard arms.
 * Software Focus: You now face the final bossdynamic balancing and
zero-moment point (ZMP) control. Because your CAN FD bus, motor firmware, and
state machines were battle-tested in Stages 1 and 2, 100% of your debugging
effort can go into the control theory rather than chasing hardware ghosts.