Pragmatic Heterogeneous Collaborative Perception via Generative Communication Mechanism

TL;DR

This paper introduces GenComm, a novel generative communication mechanism enabling heterogeneous multi-agent perception collaboration without retraining core modules, significantly reducing computational costs and supporting scalable integration of new agents.

Contribution

The paper proposes a non-intrusive, generative approach for heterogeneous multi-agent perception that preserves semantic consistency and allows efficient addition of new agents with minimal computational overhead.

Findings

GenComm outperforms state-of-the-art methods on multiple datasets.

Achieves 81% reduction in computational cost and parameters for new agents.

Effectively maintains semantic and spatial alignment across heterogeneous agents.

Abstract

Multi-agent collaboration enhances the perception capabilities of individual agents through information sharing. However, in real-world applications, differences in sensors and models across heterogeneous agents inevitably lead to domain gaps during collaboration. Existing approaches based on adaptation and reconstruction fail to support pragmatic heterogeneous collaboration due to two key limitations: (1) Intrusive retraining of the encoder or core modules disrupts the established semantic consistency among agents; and (2) accommodating new agents incurs high computational costs, limiting scalability. To address these challenges, we present a novel Generative Communication mechanism (GenComm) that facilitates seamless perception across heterogeneous multi-agent systems through feature generation, without altering the original network, and employs lightweight numerical alignment of spatial information to efficiently integrate new agents at minimal cost. Specifically, a tailored Deformable Message Extractor is designed to extract spatial message for each collaborator, which is then transmitted in place of intermediate features. The Spatial-Aware Feature Generator, utilizing a conditional diffusion model, generates features aligned with the ego agent's semantic space while preserving the spatial information of the collaborators. These generated features are further refined by a Channel Enhancer before fusion. Experiments conducted on the OPV2V-H, DAIR-V2X and V2X-Real datasets demonstrate that GenComm outperforms existing state-of-the-art methods, achieving an 81% reduction in both computational cost and parameter count when incorporating new agents. Our code is available at https://github.com/jeffreychou777/GenComm.