Roller — Balance Bot Sim

Controller

Type

Tunings

Navigation

Enable Target x (m) Target z (m) Max yaw rate (rad/s) Yaw disable radius (m) Mode Planner

Lidar A* plans on what the bot has actually seen. Drive around to fill in the map.

Kp (rad/m)

v_max (m/s) Linear zone (m) Lookahead (s)

Mixer — velocity → tilt

Closes velocity error into a body tilt. The smallest learnable layer.

Kvel (rad·s/m) Tilt limit (rad) a_max (m/s²) Mixer

Attitude — pitch (force out)

pitch_P (N/rad) pitch_D (N·s/rad) pitch_I (auto-trim) force_max (N) Pitch

RNN hyperparameters

hidden epochs seq len lr

Attitude — yaw (torque out)

heading_P (rad/s per rad) yaw_rate_max (rad/s) yaw_rate_P (N·m·s/rad) torque_max (N·m) Yaw

Wheels — force tracker

wheelbase (m) force_P (PWM/N) force_I (PWM/(N·s)) force_I_max (PWM) extra deadband (PWM)

PID gains (force out)

Kp Ki Kd Kx (cart pos) Kv (cart vel) Force limit (N)

ArduBalance — outer loop

bal_P (PWM per cd) bal_D (PWM per cd/s) bal_I (cd per cd·s) p_vel (speed fb)

ArduBalance — inner wheel PID

wheel_P wheel_I wheel_D FF PWM/mps Dead zone

Yaw control (legacy)

Used by ArduBalance, PID, and the legacy whole-stack NN. Cascade replaces this with Attitude's yaw (above).

Heading Kp (PWM per cd) Kyaw (N·m per rad/s) Max yaw torque (N·m)

Preset

Physics

Wheel+frame mass M Body mass m CoM height L (m) Wheel radius R (m) Wheel inertia Iw Rolling friction Pivot friction Yaw inertia I_yaw Yaw friction c_yaw Initial angle° Noise (rad)

Sensors & Timing

Encoder ticks/rev IMU pitch noise (rad) Gyro noise (rad/s) IMU bias drift (rad/√s) Encoder dropout (0..1) Sensor rate (Hz) Outer loop (Hz) Inner loop (Hz)

Motor

Km (N per duty) Kv (back-EMF N·s/m) PWM max Deadband (PWM)

Disturbances — make life hard

Inject impulses or biases to test disturbance rejection. A recorded-mode NN will fall here; a random-mode NN should recover.

Force impulse (N·s)

Yaw impulse (N·m·s)

IMU bias step (rad)

Shove strength (rad/s)

Top-bar Shove button kicks pitch_rate by this amount.