Physic models
1. Wheel dynamics
We introduce wheels as a mass with cylindrical shape (Figure [fig:wheel_forces]). Each wheel has following properties:
location of the wheel as to the chassis ref point [m,rad] in local coordinates \(L_w = \{ x_w, y_w, \Phi \}\)
diameter \(d_w\) [m]
width \(w_w\) [m]
mass \(m_w\) [kg]
inertia \(I_{yy}\)
spinning angular position \(\phi_w\) [rad]
spinning angular velocity \(\omega_w\) [rad/s]
Thus, each wheel is represented as \(W = \{L_w, d_w, w_w, m_w, I_{yy}, \phi_w, \omega_w\}\)
2. Friction models
Friction models base
Friction model base introduces Friction input structure, that incorporates forces of wheel
weight on this wheel from the car chassis, excluding the weight of the wheel itself \(w\) [N]
motor torque \(\tau\) [Nm]
instantaneous velocity
\[\begin{split}\nu = \begin{bmatrix} \nu_x \\ \nu_y \end{bmatrix}\end{split}\]in local coordinate frame
Default friction
At the moment, there is only one basic friction model available for vehicles. Default friction model evaluates …
Default friction evaluates forces in the wheel coordinate frame:
To calculate maximal allowed friction for the wheel, we introduce partial mass:
Where \(\mu\) is friction coefficient for wheel.
Calculating latitudinal friction (decoupled sub-problem):
Calculating wheel desired angular velocity:
Calculating longitudinal friction:
Simply composing friction forces to vector:
With new friction, we evaluate angular acceleration (code says angular velocity impulse, but the units are for acceleration) of the wheel:
Using given angular acceleration, we update wheel’s angular velocity:
Ward-Iagnemma friction
This type of friction is an implementation of paper from Chris Ward and Karl Iagnemma [WI08].
Rolling resistance is generally modeled as a combination of static- and velocity-dependant forces [17], [21]. Authors propose function with form similar to Pacejka’s model as a continuously differentiable formulation of the rolling resistance with the static force smoothed at zero velocity to avoid a singularity. The rolling resistance is
Where \(A_{roll}\), \(R_1\), \(R_2\) are the model-dependent coefficients. The impact of these coefficients is shown at figure [fig:wi_rr] taken from original paper.
This force \(F_{rr}\) is then added to \(F_{f,lon}\).
Default constants were chosen as in reference paper and showed good stability and robust results. In addition, they can be altered via configuration file.
3. Vehicle models
Vehicle models are fully configurable with world XML files.
Vehicle base class
Vehicle base incorporates basic functions for every vehicle actor in the scene. It is also responsible for updating state of vehicles.
It has implementation of interaction with world. Derived classes re-implement only work with torques/forces on wheels.
At the moment, no model takes into account the weight transfer, so weight on wheels is calculated in this base class.
Vehicle base class also provides ground-truth for velocity and position.
Before time step:
Update wheels position using Box2D
Invoke motor controllers (reimplemented in derived classes)
Evaluate friction of wheels with passed friction model
Apply force to vehicle body using Box2D
Time step - update internal vehicle state variables \(q\) and \(\dot{q}\)
After time step - updates wheels rotation
Center of mass is defined as center of Box2D shape, currently there is no +Z mobility.
Differential driven
A differential wheeled robot is a mobile robot whose movement is based on two separately driven wheels placed on either side of the robot body. It can thus change its direction by varying the relative rate of rotation of its wheels and hence does not require an additional steering motion.
Odometry-based velocity estimation is implemented via Euler formula (consider revising, it doesn’t include side slip):
Where \(\omega_{veh}\) is angular velocity of the robot, \(R_i\) - radius of the wheel, \(y_i\) is the y position of the wheel, \(\omega_i\) is the angular velocity of the wheel. All calculations in the robot’s local frame.
Nothing more interesting here.
Ackermann driven
Ackermann steering geometry is a geometric arrangement of linkages in the steering of a car or other vehicle designed to solve the problem of wheels on the inside and outside of a turn needing to trace out circles of different radii.
Ackermann wheels’ angles are computed as following:
where \(\alpha\) is the desired equivalent steering angle, \(w\) is wheels distance and \(l\) is wheels base. Outer and inner wheel are defined by the turn direction.
Odometry-based velocity estimation is implemented via Euler formula (consider revising, it doesn’t include side slip):
where \(\omega_{veh}\) is angular velocity of the robot, \(R_{ri}\) - radius of the rear wheel, \(y_{ri}\) is the y position of the rear wheel, \(\omega_{ri}\) is the angular velocity of the rear wheel. All calculations in the robot’s local frame.
Ackermann-driven with drivetrain
This type of dynamics has the same geometry as simple Ackermann-driven robots. However, its powertrain is completely different.
Instead of one “motor” per wheel, this type of dynamics incorporates one “motor” linked to wheels by differentials.
There are two types of differentials:
Open differntial
Torsen-like locking differential [Cho88]
Each type of differential can be linked with following configurations:
Front drive
Rear drive
4WD
Split is customizable between all axes.
As engine plays controller, whose torque output is then fed into differentials.
For open differential act the following equations:
Where \(K_{s,f}, K_{s,frl}, K_{s,rrl}\) are split coefficients between axes.
Different things happen for Torsen-like differentials. As this type is self-locking, its torque output per wheel depends on wheel’s velocity. Here is the function of selecting torque on the next time step based on previous time step velocity. First, introduce the bias ratio - the ratio indicating how much more torque the Torsen can send to the tire with more available traction, than is used by the tire with less traction. This ratio represents the “locking effect” of the differential. By default, it is set to \(b = 1.5\)
\(\omega_1, \omega_2\) and \(t_1, t_2\) are the output axles angular velocities and torque splits respectively. \(K_s\) is differential split when it is not locked.
Torque delivery for 2WD is pretty straightforward. There is one input from “motor” and two outputs to wheels, so wheel torques are:
where \(t_i\) is the output of Torsen differential for \(i\)-th wheel.
With 4WD, torque is first split with Torsen to front and rear parts, each of them is than split independently with another Torsen.
At the moment, there is no model of the engine and thus no feedback of tires torque to engine.
4. Controllers
Different vehicles have different controllers. At the moment, differential and Ackermann drives have their own controllers.
Controllers are divided into several types:
Raw forces
Twist
Ackermann has controller, which controls steering angle and speed.
Controllers’ input and output are described by dynamics’ classes that they use.
Differential raw controller
This type of controller has simple response to user’s input integrating wheel torque with each simulation frame.
Differential Twist controller
Differential twist controller uses PID regulator to control linear and angular speed of the robot.
Setpoints for \(v_r\) and \(v_l\) are calculated as following:
where \(\nu\) is desired linear velocity and \(\omega\) is desired angular velocity.
Inverted formula are suitable to get actual velocities from odometry estimates.
Then, velocity of the wheels is controlled with PID regulator.
Ackermann raw controller
As a raw differential controller, raw Ackermann controller integrates user input and sets wheel torques and steering wheel angle.
Ackermann twist controller
Ackermann twist controller uses PID regulator to control wheel torques responding to angular and linear velocity commands. Turn radius and desired steering angle are calculated:
Desired velocities for wheels are computed by rotating desired linear velocity to the steering angle. In the same way, actual velocities from “odometry” are computed.
Then, torque of separate wheels is controlled with PID regulators for each wheel.
Ackermann steering controller
Ackermann steering controller takes as input linear speed an steering angle.
Then, it executes Ackermann twist controller to control wheels’ torques.
Ackermann-drivetrain controllers
These controllers’ steering is identical to Ackermann contollers, however, their torque part is different.
These controllers’ output acts like ’engine’ for drivetrain. Instead of separate outputs to wheels, it has one torque output to differentials that will split it to separate wheels.