Kinematic Bicycle Model

Abstract

This repository contains a implementation of the Kinematic Bicycle model. The model can be defined with the following state-space representation,

$$ \frac{d}{dt} \begin{pmatrix} x \\ y \\ \theta \\ v \end{pmatrix} = \begin{pmatrix} v\cos{\theta} \\ v\sin{\theta} \\ \frac{v\tan{\delta}}{L} \\ a \end{pmatrix} $$

where $v$ is the vehicle's velocity in the x-axis, $\theta$ is the vehicle's yaw, $\delta$ is the steering angle, $L$ is the vehicle's wheelbase and $a$ is the acceleration/throttle. You may read more about the Kinematic Bicycle Model here.

Installation

pip install git+https://github.com/winstxnhdw/KinematicBicycleModel

Usage

The model is written entirely in Cython, completely typesafe, and drops the Global Interpreter Lock (GIL) whenever possible. The following code sample demonstrates how to use the model.

Warning

This model is not picklable as it relies on __cinit__ for fast initialisation.

from kbm import KinematicBicycleModel

model = KinematicBicycleModel(wheelbase=2.96, max_steer=0.57596, delta_time=0.05)
state = model.compute_state(
    x=0.0,
    y=0.0,
    yaw=0.0,
    steer=0.0,
    velocity=0.0,
    acceleration=5.0
)

print(f"The vehicle is at ({state['x']}, {state['y']})")
print(f"The vehicle is facing {state['yaw']} rad")
print(f"The vehicle is steering at {state['steer']} rad")
print(f"The vehicle is moving at {state['velocity']} m/s")
print(f"The vehicle is turning at {state['angular_velocity']} rad/s")

Since the GIL is dropped, the model may take advantage of multithreaded parallelism. The following code sample demonstrates an example of this.

from concurrent.futures import ThreadPoolExecutor
from random import uniform

from kbm import KinematicBicycleModel

model = KinematicBicycleModel(wheelbase=2.96, max_steer=0.57596, delta_time=0.05)
random_steer_values = (uniform(-3.14159, 3.14159) for _ in range(1000))


def compute_state(steer: float):
    return model.compute_state(x=0.0, y=0.0, yaw=0.0, steer=steer, velocity=0.0, acceleration=5.0)


state_with_highest_angular_velocity = max(
    ThreadPoolExecutor().map(compute_state, random_steer_values),
    key=lambda state: state["angular_velocity"],
)

Limitations

Just like with all other bicycle models, this model is a discrete model and loses its accuracy when the time step is set too large or the vehicle is made to travel at unreasonably high speeds. Usually, the FPS of the simulation should be set to the highest possible value for the greatest accuracy. However, for rendering high-quality GIFs, 50 FPS is found to be most optimal. See the GIF89a specification.

Demo

Recursively git clone the repository

git clone --recursive https://github.com/winstxnhdw/KinematicBicycleModel.git

Play the animation

uv run animate.py