The Micro Wheeled-Leg Robot is a fully open-source, desktop-scale balancing robot that combines two brushless wheel motors with two servo-actuated legs to achieve dynamic self-balancing locomotion. Designed to be approachable for hobbyists, students, and robotics researchers alike, it demonstrates advanced control theory — specifically Linear Quadratic Regulator (LQR) balance control — in a compact, reproducible platform that sits comfortably on a desk. Every component, from the mechanical CAD files to the PCB Gerbers to the Arduino firmware, is freely available and designed to be built at home or with low-cost fabrication services.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/MuShibo/Micro-Wheeled_leg-Robot/llms.txt
Use this file to discover all available pages before exploring further.
What Makes It Unique
The Micro Wheeled-Leg Robot claims the title of the world’s smallest desktop-class dual-wheeled-leg balancing robot. Unlike single-axis balancing robots that only roll forward and backward, this robot uses a pair of FEETECH STS3032 bus servos to dynamically raise and lower its legs, enabling variable ride height and even a jump maneuver. The entire control stack — IMU sensing, LQR balance, yaw steering, leg actuation, and browser-based remote control — runs on a single ESP32 microcontroller with no companion computer required.Build Guide
Step-by-step assembly of the mechanical structure, CNC parts, carbon fiber panels, and 3D-printed components.
Firmware Setup
Flash the ESP32, configure SimpleFOC motor parameters, calibrate the servos, and set your WiFi mode.
Operation
Power-on sequence, connecting to the robot’s WiFi hotspot, and driving with the browser joystick interface.
Software Reference
Detailed reference for firmware modules, control loop parameters, WebSocket protocol, and the Wrobot struct.
How It Works
The robot’s behavior emerges from several tightly integrated subsystems that all execute inside a single Arduino-style loop on the ESP32. ESP32 as the sole compute platform. The ESP32-WROOM-32 runs the complete control pipeline — sensor reading, balance computation, motor output, servo actuation, and WiFi communication — in one continuous main loop. There is no RTOS task separation; every subsystem gets called sequentially each iteration:bat_check(), web_loop(), mpu6050.update(), lqr_balance_loop(), yaw_loop(), leg_loop(), followed by motor FOC updates.
SimpleFOC for brushless wheel motors. Two BLDC motors, each driven by an L6234PD013TR three-phase gate driver, are controlled using the SimpleFOC library in torque/voltage control mode. Each motor has a dedicated AS5600 magnetic encoder on its own I²C bus (I2Cone on pins 19/18, I2Ctwo on pins 23/5) providing shaft angle and velocity feedback at 400 kHz.
LQR algorithm for pitch balance. Self-balancing is achieved with a Linear Quadratic Regulator decomposed into four proportional gains acting on pitch angle, pitch rate, wheel displacement, and wheel speed. The combined output LQR_u is split equally between the two motors with opposite sign to produce a net forward/backward torque that keeps the robot upright. A small integrating PID (pid_lqr_u) compensates for low-torque nonlinearity, and a center-of-gravity adaptive term (pid_zeropoint) continuously trims the balance point.
FEETECH STS3032 servos for variable leg height. Two STS3032 smart bus servos (IDs 1 and 2, connected via Serial2 at 1 Mbaud) control leg elevation. Their positions are computed from the WebSocket height command and a roll-axis correction term (pid_roll_angle) derived from MPU6050 roll data, keeping the chassis level on uneven terrain. A dedicated jump routine can command both servos to their mechanical limits at maximum speed to launch the robot briefly off the ground.
MPU6050 IMU for angle and rate feedback. An MPU6050 module, sharing the second I²C bus with the right encoder, provides pitch angle (getAngleY()), pitch rate (getGyroY()), roll angle (getAngleX()), yaw angle (getAngleZ()), and yaw rate (getGyroZ()) after automatic gyro offset calibration at startup.
WiFi + WebSocket browser remote control. The ESP32 hosts an HTTP server on port 80 serving an embedded single-page application (stored in basic_web.h) and a WebSocket server on port 81. The browser renders a joystick and control buttons; user inputs are encoded as JSON and sent over WebSocket. The firmware’s RobotProtocol::parseBasic() method decodes each JSON message into the Wrobot struct — JSON fields joy_y, joy_x, height, dir, and stable map to struct fields joyy, joyx, height, dir, and go respectively — and the control loops read directly from this struct. Both AP mode (the robot creates its own hotspot named WL…) and STA mode (the robot joins an existing network) are supported.
Open-Source Contributors
The Micro Wheeled-Leg Robot was created by Mu Shibo and Li Yufeng. The project is released under the MIT License, welcoming contributions, forks, and derivative designs from the community.All design files are available on GitHub: STL files for 3D-printed parts, CNC drawings for the steel base and aluminum motor mounts, carbon fiber panel DWG/PDF layouts, PCB schematics and Gerbers (designed in LCEDA) for all four boards, a complete BOM, and the full Arduino firmware. No files are paywalled or hidden.