Chance-Constrained Collision Avoidance Based Motion Planning in a Cooperative Perception Framework

Perceiving the environment in complex driving scenarios is critical for the safety of autonomous vehicles. Recent advancements in multiagent perception and vehicle-to-vehicle (V2V) technologies have enabled Autonomous Vehicles (AVs) not only to exchange basic safety messages but also to share their perception output with other vehicles. The concept of cooperative perception primarily addresses the challenge of dealing with occluded objects. Currently, a significant portion of research within the cooperative perception domain is dedicated to improving V2V communication systems and perception modules. However, there is a noticeable oversight in incorporating these technologies into like motion planning and vehicular control modules. A recent paper on motion planning for AVs in the presence of occlusions discusses an Optimal Control Problem (OCP) formulation using infrastructure sensor information, in which external data, coupled with the AV's individual perception, is used to determine perception reliability. This is subsequently used to estimate collision risk, ensuring that the estimated collision risk remains within the bounds of an acceptable maximum residual risk. However, the paper relies on numerous assumptions and heuristics to define parameters for the risk model and behavior options, respectively. Furthermore, delves into motion planning based on cooperative perception. However, it fails to consider the uncertainties linked to the perception outputs of neighboring AVs. There are two types of uncertainties in perception: aleatoric and epistemic uncertainty. Aleatoric uncertainty (data) is related to the characteristics of the sensors and effect of environmental conditions on their functionality. Epistemic uncertainty (model) is associated with the training of Deep Neural Networks and their ability to generalize to out-of-distribution (OOD) data. In a typical AV pipeline, the latter planning and control modules operate on processed environmental perception data. Thus, uncertainty in the outputs of implemented DNNs subsequently impact the decision-making process of AVs. The project proposal aims to address the above issues by constructing an OCP with probabilistic constraints [5,6]. Interaction-Aware motion prediction models capture the dependence between the predicted target vehicle trajectory and surrounding vehicles. Utilizing this in conjunction with perception uncertainties in OCP constraints while balancing risk and progress allows for non-conservative planning. The research team will explore convex approximations for these potentially non-convex constraints. The team may also consider investigating motion planning using reinforcement learning in case of inaccurate vehicle dynamics model. The team will also investigate False Data Injection in the framework, where an attacker may introduce fabricated data into the Ego's object detection algorithm, potentially causing collisions. The goal is to develop resilient strategies against these attacks. The algorithm's performance will be evaluated in two uncontrolled intersection scenarios: one involving a left turn collision scenario where a vehicle on the right side of the Ego is occluded, and the other focusing on planning around an occluded static obstacle. The team will mainly focus on V2V technology, with vehicles directly communicating the dynamic data required, but the team may also consider V2I and infrastructure aided cooperative perception if time permits. The team will evaluate the cooperative perception framework on OPV2V and V2V4Real datasets and in simulation environments.

• Record URL:
  https://utc.engineering.osu.edu/chance-constrained-collision-avoidance-based-motion-planning-cooperative-perception-framework

Language

• English

Project

• Status: Active
• Funding: $109877
• Contract Numbers:

  69A3552348327

• Sponsor Organizations:

  Office of the Assistant Secretary for Research and Technology

  University Transportation Centers Program
  Department of Transportation
  Washington, DC  United States  20590
• Managing Organizations:

  Center for Automated Vehicle Research with Multimodal Assured Navigation

  Ohio State University
  Columbus, OH  United States  43210
• Project Managers:

  Kline, Robin

• Performing Organizations:

  Ohio State University Center for Automotive Research

  930 Kinnear Road
  Columbus, OH  United States  43212
• Principal Investigators:

  Redmill, Keith

• Start Date: 20231030
• Expected Completion Date: 20230830
• Actual Completion Date: 0
• USDOT Program: University Transportation Centers Program

Subject/Index Terms

• TRT Terms: Algorithms; Autonomous vehicles; Computer security; Crash avoidance systems; Uncertainty; Vehicle to vehicle communications
• Subject Areas: Data and Information Technology; Highways; Security and Emergencies; Vehicles and Equipment;

Filing Info

• Accession Number: 01901376
• Record Type: Research project
• Source Agency: Center for Automated Vehicle Research with Multimodal Assured Navigation
• Contract Numbers: 69A3552348327
• Files: UTC, RIP
• Created Date: Dec 4 2023 5:10PM