Project: #573 Avoiding Collisions in Connected and Autonomous Driving Using Safe Deep Reinforcement Learning Exploration through Control Barrier/Lyapunov Functions Progress Report - Reporting Period Ending: April 1, 2026 Principal Investigator: Bilin Aksun-Guvenc Status: Active Start Date: July 1, 2025 End Date: June 30, 2026 Research Type: None Grant Type: Research Applied Grant Program: US DOT BIL, Safety21, 2023 - 2028 (4811) Grant Cycle: Safety21 : 25-26 Progress Report (Last Updated: April 11, 2026, 10:29 a.m.) % Project Completed to Date: None % Grant Award Expended: None % Match Expended & Document: None USDOT Requirements Accomplishments This project focuses on developing data driven collision avoidance algorithms that learn to avoid collisions with nearby vehicles and vulnerable road users, building upon our approach to pedestrian and bicyclist safety in our Year 1 and Year 2 Safety 21 projects. A deep reinforcement learning safety algorithm with safe exploration was developed and evaluated using realistic hardware-in-the-loop (HIL) and vehicle-in-virtual-environment (VVE) platforms. We improved the Vulnerable Road User (VRU) integration into our VVE testing by using a cloud server to share information between the vehicle(s) and VRU(s) in the shared virtual environment. We have also incorporated VR headsets such that the VRU(s) and observers in the vehicle(s) can be immersed in the same virtual environment. We have also been working on handling dynamic obstacles, autonomous lane changing in highways, and autonomous handling of square and round intersections with realistic traffic. Our deep reinforcement learning based collision avoidance algorithm learns from interactions in these complex traffic environments, adapts to various traffic scenarios, and satisfies real-time performance requirements. We use a hybrid approach in that the trained deep reinforcement learning algorithm is only activated during dangerous and close interactions with other road users. Training is based on such interactions using traffic microsimulation to generate realistic training interactions automatically. We have integrated our deep reinforcement learning algorithm with control Lyapunov and control barrier functions to limit its training to feasible areas of interaction with other road actors within the road barriers, with realistic vehicle dynamic model constraints as compared to the oversimplified ones used in the literature. While mainly one doctoral student worked as a graduate research assistant funded by the project, two other doctoral students, an undergraduate student in the Department of Mechanical and Aerospace Engineering and two masters students in the Department of Electrical and Computer Engineering of the Ohio State University also took part partially in the project research activities, thus, receiving research training and professional development support. Project results were disseminated in 1 journal paper (published), 1 conference paper (accepted), one Safety 21 Faculty Meeting presentation, one poster at the Safety 21 Deployment Partner Consortium Symposium, one poster at the OSU AI summit, one OSU Denman Research Forum poster, one OSU MAE B.S. Honor’s Thesis presentation and defense, one OSU MAE Graduate Research Showcase poster, and one OSU College of Engineering Welcome Back Exhibition day poster. During the next reporting period, we will further develop and demonstrate our results using HIL and VVE evaluations. Our dissemination activities will continue with one B.S. Honor’s thesis, the final project report, and journal and conference paper submissions. The accepted SAE WCX paper will be presented in Detroit during mid-April to academia and members of the automotive industry. Impacts As part of transportation workforce development, one doctoral graduate student took part directly while two other doctoral graduate students, two masters students and one undergraduate student took part partially in project work. The first graduate student was supported as a GRA and a second one was supported partially during the summer. These six students were trained in research on VRU integration into VVE using cloud servers, use of VR headsets, provably safe connected and autonomous driving decision making using deep reinforcement learning and control barrier functions, intersection safety applications, and HIL and VVE simulations. We have also recruited a new undergraduate student in Computer Science who will start taking part in the training activities. The OSU MAE undergraduate student mentioned above will finish his B.S. studies in May 2026 and will start working in a major automotive OEM in Detroit. One SAE WCX 2025 paper was accepted for publication and presentation with high review marks and will be presented by the doctoral students mid-April in Detroit. The SAE meeting is well attended by the automotive and transportation industries. One journal paper was published in Sensors (impact factor 3.5). Project findings were shared with our deployment partner TRC (Transportation Research Center). We also prepared a one page flyer type newsletter to share our progress with potential future industry partners. It has also been noted that the VVE approach of our research group has started receiving more attention from national labs and the vehicle testing industry. Co-PI L. Guvenc is currently teaching the ME 8322 Vehicle System Dynamics and Control and is ECE 5553 Autonomy in Vehicles courses this Spring 2026 semester. Both courses treat the safety of road transport and help with the transportation workforce development goal of the UTC. Co-PI L. Guvenc also taught the ME 8352 Robust Control of Mechatronic Systems course in the Autumn 2025 semester, and this course is related to robust and delay tolerant controls and hence also helps with transportation workforce development. Co-PI L. Guvenc also taught ME 3360 System Integration and Control during Autumn 2025 which has helped in recruiting undergraduate students. Other Courses: ME 8322 Vehicle System Dynamics and Control ECE 5553 Autonomy in Vehicles ME 8352 Robust Control of Mechatronic Systems ME 3360 System Integration and Control ECE 5553 Autonomy in Vehicles (online section attended by three GM engineers) has seen considerable revision during the current Spring 2026 semester with significantly revised lecture notes supported by many examples with their Matlab m-files. The Vulnerable Road User tracking and communication parts of the VVE approach have been updated as well using the MATLAB cloud server and the MATLAB mobile toolbox (iOS phones are also supported now) and the results/code will be shared on a GitHub site in the final report. Narrated YouTube videos illustrating some of the research results have also been prepared and the links will be shared in the final report. Outcomes New Partners We have been in contact with a local automotive supplier (several meetings have already taken place) and hope to add them as a deployment partner. Issues There have been some issues related to latencies, communication between vehicle and VRU systems during VVE, and some problems in HIL and vehicle implementations. They are all resolved or being resolved without any significant impact.