Evaluating Autonomous Urban Perception and Planning in a 1/10th Scale MiniCity

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision for Autonomous Systems

Background:

• Autonomous vehicles require robust perception systems for safe navigation.
• Evaluating perception hardware and software in realistic urban environments is challenging.

Purpose of the Study:

• To introduce MiniCity, a novel 1/10th scale urban evaluation platform for autonomous vehicle perception.
• To demonstrate MiniCity's utility in assessing perception algorithms and sensor configurations.

Main Methods:

• Developed a 1/10th scale realistic urban environment with intersections and autonomous vehicles.
• Integrated state-of-the-art sensors and algorithms onto the scaled vehicles.
• Designed urban driving scenarios, including occluded intersections and multi-vehicle interactions.

Main Results:

• Successfully evaluated perception algorithm performance for object detection and localization.
• Assessed the impact of different sensor and algorithm configurations on perception accuracy.
• Demonstrated MiniCity's capability to evaluate downstream tasks like collision avoidance and lane following.

Conclusions:

• MiniCity provides a versatile and scalable platform for rigorous testing of autonomous vehicle perception systems.
• The platform facilitates the evaluation of both perception tasks and their influence on vehicle control.
• MiniCity enables systematic assessment of sensor and algorithm trade-offs in complex urban driving scenarios.