[Paper Review] Adaptive Communications in Collaborative Perception with Domain Alignment for Autonomous Driving

Hu Senkang, Zhengru Fang+5 more|arXiv (Cornell University)|Sep 15, 2023

Visual Attention and Saliency Detection4 citations

TL;DR

This paper proposes ACC-DA, a channel-aware collaborative perception framework that dynamically optimizes communication graphs to minimize transmission delay, adapts data reconstruction for rate-distortion efficiency and reduced temporal redundancy, and employs frequency-domain domain alignment to mitigate data heterogeneity across vehicles. The method improves perception accuracy by up to 3 percentage points on vehicle detection and reduces bitrate by 111.1% at high PSNR, outperforming state-of-the-art methods in autonomous driving scenarios.

ABSTRACT

Collaborative perception among multiple connected and autonomous vehicles can greatly enhance perceptive capabilities by allowing vehicles to exchange supplementary information via communications. Despite advances in previous approaches, challenges still remain due to channel variations and data heterogeneity among collaborative vehicles. To address these issues, we propose ACC-DA, a channel-aware collaborative perception framework to dynamically adjust the communication graph and minimize the average transmission delay while mitigating the side effects from the data heterogeneity. Our novelties lie in three aspects. We first design a transmission delay minimization method, which can construct the communication graph and minimize the transmission delay according to different channel information state. We then propose an adaptive data reconstruction mechanism, which can dynamically adjust the rate-distortion trade-off to enhance perception efficiency. Moreover, it minimizes the temporal redundancy during data transmissions. Finally, we conceive a domain alignment scheme to align the data distribution from different vehicles, which can mitigate the domain gap between different vehicles and improve the performance of the target task. Comprehensive experiments demonstrate the effectiveness of our method in comparison to the existing state-of-the-art works.

Motivation & Objective

• To address dynamic channel variations and data heterogeneity in multi-vehicle collaborative perception for autonomous driving.
• To minimize transmission delay by constructing communication graphs based on real-time channel state information (CSI).
• To improve transmission efficiency through adaptive rate-distortion trade-off and real-time model refinement in data reconstruction.
• To reduce domain gaps caused by sensor diversity and environmental differences across vehicles.
• To enhance joint perception performance by aligning feature distributions across vehicles using frequency-domain amplitude spectrum alignment.

Proposed method

• A transmission delay minimization module constructs dynamic communication graphs using real-time CSI to reduce average delay.
• An adaptive data reconstruction mechanism adjusts the rate-distortion trade-off in real time and minimizes temporal redundancy via online model refinement.
• A domain alignment scheme transforms images to the frequency domain and aligns their amplitude spectra to reduce distribution heterogeneity.
• The framework integrates three modules: delay-aware communication, adaptive reconstruction, and domain alignment, for end-to-end collaborative perception.
• The method uses t-SNE visualization to validate reduced data distribution divergence after domain alignment.
• It leverages existing perception models (e.g., V2VNet, DiscoNet, Attention Fusion) as baselines and enhances them with the proposed modules.

Experimental results

Research questions

• RQ1 How can communication graphs be dynamically optimized to minimize transmission delay under varying wireless channel conditions in vehicular networks?
• RQ2 To what extent can adaptive rate-distortion trade-off and real-time model refinement improve data reconstruction efficiency and reduce redundancy in collaborative perception?
• RQ3 Can frequency-domain amplitude spectrum alignment effectively reduce data heterogeneity caused by sensor differences and environmental variations across vehicles?
• RQ4 How does domain alignment improve joint perception performance in collaborative autonomous driving systems?
• RQ5 What is the combined impact of delay minimization, adaptive reconstruction, and domain alignment on overall collaborative perception accuracy and efficiency?

Key findings

• The domain alignment mechanism improved vehicle detection accuracy by 1.38% for Attention Fusion, 2.10% for DiscoNet, and 1.82% for V2VNet on the OPV2V dataset.
• With domain alignment, the overall AP@IoU increased from 54.08% to 55.06% for the proposed ACC-DA framework.
• The adaptive refinement strategy achieved a 25% bitrate reduction at 38.0 dB PSNR and a 111.1% reduction at 38.6 dB PSNR compared to non-refined reconstruction.
• t-SNE visualization confirmed that domain alignment reduced distribution heterogeneity, with transformed features showing tighter clustering than original data.
• The transmission delay minimization module effectively reduced average delay by adapting the communication graph to dynamic CSI.
• Comprehensive experiments demonstrated that ACC-DA outperforms state-of-the-art methods in both perception accuracy and communication efficiency.