Multi-agent Collaborative Perception for Autonomous Driving:
Unsettled Aspects EPR2023017
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:
Establishing normative
communication protocols to facilitate seamless information sharing
among vehicles
Defining collaboration
strategies, including identifying specific collaboration projects,
partners, and content, as well as establishing the integration
mechanism
Collecting sufficient data
for MCP model training, including capturing diverse modal data and
labeling various downstream tasks as accurately as
possible
