Holonomic Control

Following snippets shows how to control a robot with omnidirectional wheels. Holonomic wheels are wheels with 2 degrees of freedom. They are also known as omni-directional drive wheels, omni-wheels or omniwheels which ever floats your boat.

Omniwheels allows movement in any direction, at any angle, without rotating beforehand. For an omniwheel robot to translate at a certain angle, each motor needs to go at a certain speed in relation to the other motors. Speed doesn't matter, just the ratios and to rotate at some particular speed, we must add or subtract equally from the motor speed of each motor.

These ratios can be calculated using the following formula assuming n ≥ 3,

\begin{equation} \begin{bmatrix} v_1&v_2&v_3&v_4 \end{bmatrix}^T = \begin{bmatrix} -sin\theta_1&cos\theta_1&1\\ -sin\theta_2&cos\theta_2&1\\ .. & .. & .. \\ -sin\theta_n&cos\theta_n&1\\ \end{bmatrix} \times \begin{bmatrix} v_x&v_y&R_W \end{bmatrix}^T \end{equation}

The idea is really simple, in a nutshell each motor pushes the robot in two directions X and Y (their sin and cos components). Basically we are looking for the combination of motor speeds v1, v2, v3, v4, that when combined gives us vx, vy.

All the angles of the motor axis are measured relative to the x direction in the coordinate system of the robot. Each rotating the robot in counter counterclockwise direction.

(def *wheel-angles* [60 135 225 300])

The intended direction of travel of the robot is calculated by the control system in terms of Vx, Vy and Rw.

(defn velocity-coupling-matrix [& n]
  (reduce (fn[h v]
            (conj h [(- (Math/sin (Math/toRadians v)))
                     (Math/cos (Math/toRadians v)) 1])) [] n))

(defn motor-ratios [vcm x y r]
  (flatten (matrix-seq (matrix-multiply (matrix vcm) (matrix [[x][y][r]])))))

Multiplying velocity coupling matrix with our intended direction of travel results in a sequence of motor speed ratios.

At this point you have couple of options for movement,

(let [vcm (apply velocity-coupling-matrix *wheel-angles*)
      base-rps 300]
  (defn motor-speeds [{x :x y :y} r]
    (map #(* base-rps %) (motor-ratios vcm x y r))))

Calculate a normalized vector to your target and multiply that with a base motor speed.

(def *max-velocity* 5)
(def *slowing-dist* (* 5 *robot-width*))

(defn arrive [self target velocity]
  (let [to-target (- target self)
        dist (magnitude to-target)
        ramped-speed (* *max-velocity* (/ dist *slowing-dist*))
        clipped-speed (min ramped-speed *max-velocity*)
        desired-velocity (* to-target (/ clipped-speed dist))]
    (- desired-velocity velocity)))

Or calculate proper velocity that will get you to your target. Above is the arrive behavior from Steering Behaviors For Autonomous Characters, you can then convert that velocity to motor speeds using the following formula, assuming your velocity is in \(\frac{meter}{sec}\) you can convert it to \(\frac{radian}{sec}\) using,

\begin{equation} W = \frac{2 \times v}{d} \end{equation}


  • w -> angular speed (\(\frac{radian}{sec}\))
  • v -> linear speed (\(\frac{meter}{sec}\))
  • d -> diameter (meter)
(def *wheel-diameter* 0.054)

(let [vcm (apply velocity-coupling-matrix *wheel-angles*)
      angular-velocity #(/ (* 2 %) *wheel-diameter*)]
  (defn motor-speeds [{x :x y :y} r]
    (motor-ratios vcm
                  (angular-velocity x)
                  (angular-velocity y)
                  (angular-velocity r))))

One thing that will bite you if you are not careful is that as the robot's orientation changes, orientation of the robot's x axis also changes, so the controller needs to account for that.

Matrix operations (from commons-math),

(defn matrix [seq]
  (Array2DRowRealMatrix. (into-array (map double-array seq))))

(defn matrix-multiply [a b]
  (.multiply a b))

(defn matrix-seq [m]
  (map seq (.getData m)))