Edge-Coordinated On-Road Perception for
Connected Autonomous Vehicles Using Point
Cloud
IV. EXPERIMENTS AND RESULTS
I. INTRODUCTION The experiments were conducted on an urban driving
dataset (OPV2V) under various traffic and bandwidth
The paper explores the challenge of selecting the most
conditions. The experiments compared the proposed
valuable connected autonomous vehicles (CAVs) to provide
algorithm with several benchmarks:
point cloud data (PCD) for edge-coordinated on-road
perception. The goal is to improve the perception capabilities 1. Bandwidth-Thrift: Employs only one CAV with
of edge nodes (ENs) by utilizing the data collected from the most utility gain to conserve bandwidth.
CAVs while addressing computational and bandwidth
constraints. The paper provides insights into how to manage 2. Unlimited: Uses all available CAVs without
CAV contributions to achieve optimal road environment considering latency constraints, leading to the
understanding. highest perception performance.
3. Agnostic: Selects CAVs randomly without
II. KEY INSIGHTS AND PROBLEM optimizing for utility.
The generated PCD size grows linearly with the number 4. Area-Maximum: Selects CAVs based on
of CAVs, which strains the system’s computational and maximizing their sensing range rather than utility.
communication resources. The mobility of the CAVs further
complicates the situation, as the quality and usefulness of Under normal network conditions, the proposed method
their PCD data vary based on traffic conditions and their outperformed the other benchmarks, achieving a high
movement. probability of detecting over 80 vehicles. The method also
showed superior performance under constrained network
Given this, the paper formulates an optimization problem conditions, such as reduced bandwidth. Even though the
that seeks to maximize the utility of the EN while Unlimited method performed slightly better in perception, it
minimizing latency. The utility refers to the quality of road failed to meet the latency requirements, which are crucial for
perception achieved by integrating PCD from CAVs. The real-time applications. The proposed method kept latency
challenge is selecting the right subset of CAVs to employ for within 0.1 seconds, making it more suitable for real-time
data collection, considering that employing too many CAVs perception in dynamic traffic environments.
at once results in redundant information and increased
processing delays V. CONCLUSION
The paper concludes that the proposed method strikes an
optimal balance between computational efficiency and
III. PROPOSED SOLUTION perception performance. By selectively employing CAVs,
the system can achieve high-quality road perception while
The paper proposes a CAV selection algorithm that keeping within latency and bandwidth limits. The research
prioritizes vehicles based on their marginal contribution to provides a robust solution to improve edge-coordinated
the overall perception. The algorithm evaluates the gain in perception in autonomous driving, particularly in urban
utility from each CAV and selects the subset that maximizes environments where traffic conditions and road layouts are
road perception while keeping within the bandwidth and highly dynamic.
latency constraints. The key objective is to ensure that the
system achieves a balance between computational efficiency
and perception accuracy. VI. PROPOSALS
This algorithm is designed to solve the optimization
[1] Bandwidth-Efficient Edge-Assisted Perception for Dense Urban
problem using a greedy approach. In each time slot, the EN Environments.
selects CAVs based on their potential marginal gain in [2] Latency-Aware Collaborative Perception in Autonomous Vehicles
utility, aiming to provide the best possible perception of the [3] Adaptive Point Cloud Data Compression for Real-Time Perception.
road while adhering to a set latency constraint. The process
continues over multiple time slots, with the EN dynamically
adjusting the selection of CAVs based on changing road
conditions and the mobility of the vehicles.
XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE
