Misbehavior Detection in Uncertainty Aware Object Level Cooperative Perception for Motion Planning

Details

• Creators:
  Redmill, Keith A ; Srinivasan, Vignesh
• Corporate Creators:
  Ohio State University. Center for Automotive Research
• 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 ; Center for Automated Vehicles Research with Multimodal Assured Navigation (CARMEN+) Tier-1 University Transportation Center (UTC)
• Subject/TRT Terms:
  Behavior CARLA Simulation Chance Constraints Cooperatives Mathematical Prediction Misbehavior Detection Model Predictive Control Motion Multi-Object Tracking Object Spoofing Perception Simulation Tracking Systems Uncertainty Quantification Urban Transportation

  More +
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report: October 2023 to August 2025
• Corporate Publisher:
  Center for Automated Vehicles Research with Multimodal Assured Navigation (CARMEN+) Tier-1 University Transportation Center (UTC)
• Abstract:
  Cooperative perception broadens the sensing range of an automated vehicle, especially in urban settings where other road users and infrastructure create occlusions. However, it raises safety concerns because received information can be incorrect due to faults or malicious attacks, which can lead to unsafe planning. We study object-level fusion strategies and related attack models in which an attacker communicates spoofed objects or alters reported object locations. Our object-level fusion merges detections from connected vehicles using inverse variance weighted box fusion, removes highly uncertain detections, and forwards fused detections and variances to a tracker. We validate on the V2V4Real dataset and observe that the method outperforms standard late fusion and surpasses several state-of-the-art intermediate fusion schemes. We address data trust with a misbehavior detection scheme that uses overlapping fields of view to test the contextual validity of received data and to update dynamic, source-specific trust scores based on consistency. Fusion is performed only after these scores are evaluated. The scheme is tested in two scenarios, a T junction with a spoofed vehicle and a case with a communicating vehicle that has erroneous pose, and it identifies misbehaving vehicles and excludes them from fusion. We consider uncertainty-aware trajectory planning by formulating an optimal control problem solved with model predictive control with probabilistic constraints. We validate the stack in four scenarios, a left turn at a T junction with an adversarial non-communicating vehicle, a collaborative perception case with an occluded non-communicating vehicle at a T-junction, a cooperative perception case with an occluded non-communicating stopped vehicle that requires collision avoidance during a right turn at a four-way intersection, and a similar case with the addition of a spoofed vehicle.
• Format:
  PDF
• Funding:
  69A3552348327
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:aba0d3d19654165806d3e38e9dcf600fae496eac68ae2b845cc73b318f864a65460e8f3254856fe6546f1af08eca281cf7528461e34ca4e877b62bc31d206caf
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/87356/dot_87356_DS1.pdf
• File Type:
  [PDF - 12.91 MB ]