Avoiding Collisions in Connected and Autonomous Driving Using Safe Deep Reinforcement Learning Exploration Through Control Barrier/Lyapunov Functions

Details

• Creators:
  Chen, Haochong ; Cao, Xincheng ; Guvenc, Levent ; Aksun-Guvenc, Bilin
• Corporate Creators:
  Carnegie Mellon University. Traffic21 Institute. Safety21 University Transportation Center (UTC) ; Ohio State University
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  • Traffic safety
  • Vulnerable road users
  • Crash avoidance systems
  • Autonomous vehicles
  • Connected vehicles
  • Algorithms
  • Hardware in the loop simulation
  • Highways
  • Pedestrians
  • Cyclists
  More +
• Publication/ Report Number:
  573
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report
• Corporate Publisher:
  Carnegie Mellon University. Traffic21 Institute. Safety21 University Transportation Center (UTC)
• Abstract:
  This research aims to address several key challenges in road traffic safety including vulnerable road users(VRUs) by improving decision-making, vehicle control, and system evaluation of autonomous driving. The main contributions of this research are summarized as follows:(1) An HOCLF-HOCBF-QP-based low-level control strategy is proposed to simultaneously ensure accurate trajectory tracking and collision avoidance. HOBCF acts as a safety filter for deep reinforcement learning based higher level decision making and HOBCF guarantees stable response. The proposed controller improves both the safety and robustness of ego-vehicle control in complex traffic environments.(2) A hierarchical autonomous driving framework is developed by integrating a deep reinforcement learning(DRL)-based high-level decision-making agent with the proposed HOCLF-HOCBF-QP low-level controller. The framework enables safe, smooth, and efficient navigation of ego-vehicles. In addition, different DRL algorithms including SAC-D, DDQN, etc. are systematically compared and evaluated in this project using highway driving, intersection management and VRU interaction scenarios.(3) Hardware-in-the-Loop (HIL) and Vehicle-in-Virtual-Environment (VVE) approaches are used to evaluate autonomous driving systems under complex and high-risk traffic scenarios involving VRUs. VVE-based real vehicle experiments are performed to demonstrate its effectiveness for realistic ADAS and AV function evaluation and validation.
• Format:
  PDF
• Funding:
  69A3552344811 ; 69A3552348316
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/93507/dot_93507_DS1.pdf
• File Type:
  [PDF - 13.01 MB]
• Collection(s):
  Autonomous Vehicles ; University Transportation Centers
• Main Document Checksum:
  urn:sha-512:b33eb6ba13852426108f83ee4e700f921b19707904d2900cd1b47fe970a9989fdeb7bb489eac370a5caa8714d8c23a77d02e57b65f141f278f515f1b40ac4790