Roundabout Connected/Automated Vehicle Active Inference Control Strategies to Improve Safety for All Users: Phase-I

Details

• Creators:
  Morris, Ted L. ; Shenoy, Varadaraya G. ; Shepard, Ben ; Morellas, Vassilios
• Corporate Creators:
  Minnesota Robotics Institute (MnRI) ; University of Minnesota, Twin Cities. Gemini-Huntley Robotics Research Laboratory
• Corporate Contributors:
  Center for Connected and Automated Transportation (CCAT) ; United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology
• Subject/TRT Terms:
  • Autonomous vehicles
  • Blind spots
  • Connected vehicles
  • Deep learning
  • Detection and identification system applications
  • Roundabouts
  • Sensors
  • System architecture
  • Traffic safety
• Publication/ Report Number:
  CTS 26-08
• DOI:
  https://hdl.handle.net/11299/281279
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report (Feb 2024 - Feb 2026)
• Corporate Publisher:
  University of Michigan. Center for Connected and Automated Transportation
• Description:
  Autonomous and connected vehicle perception capabilities can be enhanced with roadside sensing, by increasing the spatial range of traffic events for all road users and revealing 'blind spots' where onboard vehicle sensors cannot 'see'. A multi-lane roundabout intersection can manifest such scenarios, that introduce perception and control challenges to the autonomous vehicle. This project developed and assessed a roadside sensing and detection system that can be integrated with connected and autonomous vehicle control and information sharing research studies. The project was proposed as the first of two phases, whereby the intention of the first phase was to develop the roadside traffic detection system architecture for all road users. The system consists of a single high resolution fisheye camera sensor, located above the center of an urban arterial 2-lane roundabout intersection, a wireless 5G communication link, and edge computing hardware. During the time of the project, 3 State-of-the-art Deep Learning one-shot YOLO (You Only Look Once) models were trained and evaluated for fisheye optics. Robust detection was achieved for all models, with mAP50-95 scores 0.743, 0.634, and 0.607 for multi-image, and direct single fisheye image object detection methods, across diverse environment conditions. Vehicle tracking evaluations indicated that over 70% of all vehicle trajectories are fully tracked across all directions and maneuvers; the track errors primarily occurred before vehicles entered the roundabout. Improvements to pedestrian identification and detection should be pursued specifically to better address vulnerable road user movement states within the roundabout intersection and improve overall system performance. This will benefit the planned Phase-II objective of developing and testing a probabilistic self-learning, adaptive approach to control a connected, autonomous entering vehicle through Active Inference.
• Format:
  PDF
• Funding:
  69A3551747105
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/94063/dot_94063_DS1.pdf
• File Type:
  [PDF - 4.21 MB]
• Collection(s):
  Autonomous Vehicles ; University Transportation Centers
• Main Document Checksum:
  urn:sha-512:5de534ffa31dcbf67ee270f1a8ce2756eddf018950151445f6ba6305a92bc77e21c0f866c72b0575d0af5ea84a5d99be4aaca9f7e99740299e3dc215a58fd702