Current focus: RL + sim-to-real Now: Placement · MindSpire Labs Location: UK Status: Open to PhD 2028
PRJ-02 / QUADRUPED / 12 DOF Working

Quadruped Spider Robot

My own 12-DOF quadruped on a custom power PCB. It walks and performs multiple moves with hand-designed control. Reinforcement learning and sim-to-real in MuJoCo are next.

FOOT SPAN ESP32-S3 + PCB PLAN VIEW · 12 DOF · 3/LEG

What it is

A 12-DOF quadruped I designed and built myself. It walks and performs multiple moves using hand-designed (non-learned) control. It is the hardware platform for my reinforcement learning and sim-to-real work.

What I built

  • Actuation: ST3020 and SC15 serial bus servos.
  • Control: an ESP32-S3 controller and a Jetson Orin Nano Super.
  • Custom power PCB:
    • Two servo buses split by voltage: ST3020 servos on 12 V, SC15 servos on about 8 V from a TPS54540 buck converter.
    • Input protection.
    • INA226 current sensing and ADS1115 battery monitoring.
    • Half-duplex transceivers for the serial servo buses.

What’s next

Train locomotion policies with reinforcement learning in MuJoCo and transfer them to this robot (sim-to-real). This is the core of my research direction.