Multi-agent Collaborative Perception for Autonomous Driving: Unsettled Aspects

This report delves into the field of multi-agent collaborative perception (MCP) for autonomous driving: an area that remains unresolved. Current single-agent perception systems suffer from limitations, such as occlusion and sparse sensor observation at a far distance.Multi-agent Collaborative Perception for Autonomous Driving: Unsettled Aspects addresses three unsettled topics that demand immediate attention: (1) Establishing normative communication protocols to facilitate seamless information sharing among vehicles; (2) Defining collaboration strategies, including identifying specific collaboration projects, partners, and content, as well as establishing the integration mechanism; (3) Collecting sufficient data for MCP model training, including capturing diverse modal data and labeling various downstream tasks as accurately as possible.

• Record URL:
  • https://doi.org/10.4271/EPR2023017
• Supplemental Notes:
  • Abstract reprinted with permission of SAE International.
• Authors:
  • Chen, Guang
• Publication Date: 2023-8-15

Language

• English

Media Info

• Media Type: Web
• Features: References;
• Serial:
  • SAE EDGE™ Research Reports
  • Publisher: SAE International
  • Serial URL: https://www.sae.org/publications/edge-research-reports

Subject/Index Terms

• TRT Terms: Actuators; Autonomous vehicles; Computer network protocols; Cooperation; Education and training; Partnerships; Sensors
• Subject Areas: Highways; Vehicles and Equipment;

Filing Info

• Accession Number: 01890374
• Record Type: Publication
• Source Agency: SAE International
• Report/Paper Numbers: EPR2023017
• Files: TRIS, SAE
• Created Date: Aug 22 2023 3:30PM