Ball Balancing Robot

Moving a ball along trajectories using computer vision, PID control, inverse kinematics, and three stepper motors.

triangle_trajectory.mp4

Architecture

Package	Input	Details	Output
computer_vision	camera frames (USB)	compute ball position using Circle Hough Transform	ball (x, y) in global frame
trajectory_planner	ball (x, y)	stores and updates target waypoints	current target position
pid	ball (x, y) and target (x, y)	calculates tilt adjustments to minimize positional, accumulated, velocity errors	direction and angle of platform tilt
inverse_kinematics	direction and angle of platform tilt	convert tilt target to stepper motor angles	angle for each stepper motor
py2motor	3 stepper motor angles	communicate with stepper driver over serial	instructions encoded with custom protocol
core	static config specifying trajectory and controller params	integration of modules, sanity checks between packages	None
serial2py	joystick position over serial	convert joystick commands of target tilt positions (used to test subsystems)	direction and angle of platform tilt

flowchart LR
    subgraph core.main
        serial2py.read_arduino_joystick --> inverse_kinematics
        computer_vision --> pid
        computer_vision --> motion_planner
        motion_planner --> pid
        pid --> inverse_kinematics --> py2motor
    end

Loading

Installation

1. Install Dependencies:

   pip install -e src

2. Start the Application:

   start

To change the mode of operation, the static config can be changed in src/core/main.py

Since serial interfaces may vary between devices, they may need to be changed.

Reproducing the Setup

Required Hardware

• Camera: Connected to the device running core.main (e.g. Raspberry Pi)
• Transparent plate: (polycarbonate)
• Three linkages: Connecting the plate to the motors
• Three motors
• Motor driver: Implements the protocol specified below

Embedded Protocol

All commands should be ASCII encoded and insensitive to blank spaces.

On startup, the motor driver should receive a <h> command via serial, raise the platform to an acceptable starting position, and emit a HOME command via serial in response.

The driver should accept commands of the format <a,rad1,rad2,rad3> representing an absolute movement of the motors to a given degree in radians.

Computer Vision

The computer_vision module can be run independently with

python src/computer_vision/ball_detection.py

It may be necessary to adjust the parameters for the Hough Circle Transform to improve ball detection.