Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 150432

150432 - Developing a Cyber-Physical Intersection Testbed for Autonomous Vehicle Algorithms: Simulating Real-World Urban Traffic Dynamics Using Python 

Purpose: Safety and efficiency are key to designing and programming productive autonomous vehicles. Central to the design philosophy of these algorithms is a high level of adaptability for different environments. Thus, in this project, my task is to simulate the behavior of autonomous vehicles in a variety of urban surroundings, including a digital twin of a real world urban intersection, which in turn can be used to replicate complex urban traffic scenarios. The vehicles replicate the wheelbase and turning radius of AWS DeepRacer vehicles, which are RC cars equipped with a powerful computer to run autonomous driving algorithms, which will first be tested out on the digital testbed. Through this simulation, we are able to use the digital environment as a development testbed to lay the foundation for future research in urban autonomous vehicle algorithms.

Methods: The surrounding vehicles are simulated using the PyGame library. This allows for real time rendering of computer graphics for dynamic testing purposes. The use of a Kinematic Bicycle Model, which uses parameters such as Wheel Base, Turning Radius, Yaw rotation, Velocity and Acceleration is used to model the movement of real-world vehicles, and different parameters can be tweaked to simulate different classes of vehicles. There are two types of vehicles in the simulation; surrounding vehicles which follow predetermined paths and receive instructions periodically, and the autonomous vehicle(s) which are programmed using Model Predictive Control (MPC) built with Robot Operating System 2 (ROS2). ROS2 is an open-source state of the art framework for robotics projects, used in research and industry, and is an important layer that allows the digital vehicles to properly replicate the behavior of the real world autonomous vehicles that we will test. Surrounding vehicles will also be equipped with the Computer Vision framework, which will allow for emergency braking and turning. Central to the simulation of surrounding vehicles is the use of SimPy, which gives instructions to the surrounding vehicles on when to perform behaviors such as turning or stopping at a traffic light. This simulates a Data Distribution Service (DDS) such as eProsima DDS that allows a network of robots to receive and send data to a central computer. Lastly, different visualization methods using Pandas, NumPy, and centrally Matplotlib/Seaborn will be used to record and study the data from the vehicles in order to optimize safety and efficiency in the real world. Through the development and implementation of this digital testbed, we have found important results that have aided fellow researchers in the development of their own autonomous algorithms for complicated urban environments.

Presenting Author: David Eyal Columbia University

Presenting Author Biography: David Eyal is a sophomore at Columbia University studying Computer Science and Statistics with a passion for robotics.

Authors:

Qi Gao Columbia University
David Eyal Columbia University