Advanced Simulator for Rural AV Development

This simulator, located at the University of Tennessee, is used to enhance automated vehicle operation and safety on rural roads. The simulator was designed to provide a platform for Connected and Automated Vehicles (CAVs) deployment in rural regions. It addresses major challenges in rural areas due to unstructured roadways, limited infrastructure, and safety-critical edge cases. The project, led by researchers at the University of Tennessee, develops a comprehensive framework for enabling safe and reliable CAV operation in rural environments. Key innovations include a rural-specific perception architecture, a Perception Error Model (PEM) based on FETSGAN trained on real-world data, and GENESIS-RL, a reinforcement learning framework for generating natural yet adversarial test scenarios. Integration with vehicle-, model-, and software-in-the-loop testing—featuring a Toyota RAV4 on a dynamometer—demonstrated high-fidelity validation. PEM simulations can reveal safety metrics. GENESIS-RL can identify rural-relevant vulnerabilities like occlusion-induced delays in obstacle detection. This simulator and associated analytics establish a scalable and statistically grounded methodology for evaluating AV safety in rural contexts and provide actionable guidance for simulation protocols, infrastructure investment, and safety standards. The resulting frameworks bridge the urban-rural AV divide and support data-driven advancements in automated mobility.

In a simulation, edge cases are identified: the AV approaches a stopped vehicle on an unmarked road. Another scenario represents a vehicle executing a blind turn in front of AV. Both situations are intended to highlight the unique challenges facing self-driving technology within a rural setting.