BOLT: Online Lightweight Adaptation for Preparation-Free Heterogeneous Cooperative Perception

2026-05-01

AI Summary

This work addresses the challenge of effectively fusing perceptual information among heterogeneous autonomous agents without prior joint training or coordination. To this end, it proposes a lightweight, plug-and-play module that enables heterogeneous cooperative perception without requiring pre-deployment alignment for the first time. The method leverages the ego vehicle’s high-confidence predictions as teacher signals to guide online feature alignment of neighboring agents, while a lightweight feature adaptation and cross-agent fusion strategy enriches the ego vehicle’s low-confidence regions with complementary information. Evaluated on the DAIR-V2X and OPV2V datasets, the approach achieves up to a 32.3-point improvement in AP@50, substantially outperforming both single-agent baselines and unadapted fusion methods.

Abstract

Most existing heterogeneous cooperative perception methods depend on prior preparation like offline joint training or tailored collaborator-model adaptation. Such preprocessing is, however, generally impractical in real scenarios, as agents are usually independently trained by different developers and meet occasionally online. This work investigates \emph{preparation-free heterogeneous cooperative perception}, where agents use independently trained single-agent detectors without any pre-deployment coordination. We find direct cross-agent fusion under this setting greatly underperforms ego-only perception. We present BOLT, a lightweight plug-and-play module that adapts neighboring features online via ego-as-teacher distillation, requiring only ego predictions without ground-truth labels. BOLT leverages high-confidence ego perception features to guide cross-agent feature-domain alignment, while enabling neighbors to contribute features in the ego's low-confidence regions. With only 0.9M trainable parameters, BOLT improves AP@50 by up to 32.3 points over vanilla unadapted fusion in the preparation-free setting. It consistently outperforms ego-only results on DAIR-V2X and OPV2V, across different encoder pairs and fusion strategies. Code: https://github.com/sidiangongyuan/BOLT.

Problem

• heterogeneous cooperative perception
• preparation-free
• online adaptation
• feature fusion
• ego-as-teacher

Innovation

• preparation-free cooperative perception
• online lightweight adaptation
• ego-as-teacher distillation
• heterogeneous perception
• feature-domain alignment