Coaxial Swerve Drive From Scratch

This is easily the biggest project I've ever dedicated myself to, and although I could easily explain every last detail, I'll try to keep it concise for your sake.

Hardware Features ( ~$4000 )

• Eight Rev NEO Brushless motors (four for steer, and four for drive)

  • Mounted upside down inside the frame for better protection and lower center of gravity

• Eight Rev CAN SPARK MAX Motor Controllers to control the motors through a CAN bus interface

  • Mounted to custom 3d-printed brackets for cable management and modularity

• High torque, high speed 1:6.75 drive gear ratio

  • 14.5 ft/s -> 10 mph

• Four CTR Electronics CANcoders

  • rotary magnetic absolute encoders that communicate over the CAN bus for precise steering angle readouts

• Precision-milled skeletonized aluminum frames from Swerve Drive Specialties

• NavX AHRS IMU Gyroscope using SPI XMP RoboRIO port

• 4" Colson wheels with a coefficient of friction of 1.0

Software Features ( ~2,000 lines of code, ~15,000 total line changes )

• Teleoperated Driver Control Via Wireless Xbox Controllers

  • Custom-tuned input curves and headbands

  • Uses haptic rumble to communicate with the driver

  • Limits acceleration to prevent tipping

  • Robot retains heading even if bumped or disturbed

  • Automatically compensates for the slight curve you get when moving in a straight line while rotating

• Fully Autonomous Complex Driving

  • Fits bezier curves to a set of splines that are defined before the match to execute complex behaviors without driver input

  • Autonomous paths can be drawn on a virtual field through custom software

  • Configurable coefficient of friction to compensate for wheel slipping

  • Tuned using multiple profiled proportional–integral–derivative controllers to get sub-centimeter precision

  • Compensates for disturbances and collisions using a built-in gyroscope and vision system

    • Detects April tags to triangulate its own position in 3D space for precision movements

      • Uses Kalman filtering and linear quadratic estimation to reduce noisy results

• Advanced Motion Profiling

  • Uses Feedforward and profiled PID controllers for each module to maintain precise wheel velocities and directions

  • Uses built-in motor controllers and encoders to get 1,000 Hz update rates for sub-millisecond adjustments

  • Seeds the built-in steering encoder with the absolute CANCoder to keep continuity between the reported steering angle and the motor controllers

• Safety

  • All motors and motor controllers have a limit on current draw in Amps to reduce power spikes and retain battery and motor health

    • They also have a ramp rate limit for this same reason

  • If motor temperatures ever go beyond 65.0 °C, they are forcibly disabled until they have cooled off

  • A maximum motor power can be set when testing new code so parts aren't destroyed if something goes wrong

  • If motor controllers aren't updated every 20ms, they will stop by default

• Telemetry and Logging

  • Almost every aspect of the robot is logged at all times (20,000 data points per second)

  • The log files are stored as binary, so they have very little overhead

  • Saved to a USB stick for easy file transfers

  • Logs are timestamped, so you can rewind the logs and see exactly what went wrong

• Full Physics Simulation

  • The whole chassis can also be entirely simulated

  • The simulated wheels use flywheel physics and moment of inertia calculations along with empirical data from the motor manufacturer to produce the most realistic simulation

    • It's so accurate that motion profiling can be done entirely virtually

  • Any code, including automated and manual driving code, works exactly the same, meaning you can test and develop new software without ever touching the robot.

  • The field is also simulated, and in 3D, so you can practice driving the robot in an entirely simulated environment.

  • Because the motors are also simulated, the amount of voltage and current can also be observed over the whole system

    • This means that the strain on the battery can also be simulated, and can detect brownouts when driving

• Ease of Use

  • Almost all of the math and low-level programming is abstracted away for easy control

    • Instead of commanding each module, you only need to command the whole robot, and the software handles the rest for you

  • All the configuration is in a single file

    • Supports any frame size, wheel size, mass, acceleration, gear reduction, and much more

  • Logging is simple with an easy utility class

Further Work

I'm continuing this project with my Senior Capstone, in which I lead a course to teach and pass down my knowledge about Swerve Drive, programming, and electronics. I've essentially been the solo programmer for the robotics team, and If I don't effectively teach some of the younger members of the team what I've learned, they'll have to re-learn everything from scratch.
If you'd like to see more information on my Capstone project, my proposal is linked here