Safely Test Autonomous Vehicles with Augmented Reality

Wang, Shenlong; Forsyth, David

Permalink

https://hdl.handle.net/2142/115967 Copy

Description

Title

Safely Test Autonomous Vehicles with Augmented Reality

Author(s)

• Wang, Shenlong
• Forsyth, David

Issue Date

2022-08

Keyword(s)

• Testing
• Validation
• Simulation
• Safe Autonomy
• Photorealistic Rendering

Abstract

This work exploits augmented reality to safely train and validate autonomous vehicles’ performance in the real world under safety-critical scenarios. Toward this goal, we first develop algorithms that create virtual traffic participants with risky behaviors and seamlessly insert the virtual events into real images perceived from the physical world. The resulting composed images are photorealistic and physically grounded. The manipulated images are fed into the autonomous vehicle during testing, allowing the self-driving vehicle to react to such virtual events within either a photorealistic simulator or a real-world test track and real hardware systems. Our presented technique allows us to develop safe, hardware-in-the-loop, and cost-effective tests for self-driving cars to respond to immersive safety-critical traffic scenarios.

Publisher

Illinois Center for Transportation

Series/Report Name or Number

I-ACT-21-07

Type of Resource

text

Language

en

Copyright and License Information

No restrictions. This document is available through the National Technical Information Service, Springfield, VA 22161.

Owning Collections