Faster-HEAL: An Efficient and Privacy-Preserving Collaborative Perception Framework for Heterogeneous Autonomous Vehicles

Armin Maleki; Hayder Radha

cs.CVMar 7, 2026

Faster-HEAL: An Efficient and Privacy-Preserving Collaborative Perception Framework for Heterogeneous Autonomous Vehicles

Authors: Armin Maleki, Hayder Radha

Organizations: Electrical and Computer Engineering, Michigan State University, East Lansing, Michigan

Abstract

Collaborative perception (CP) is a promising paradigm for improving situational awareness in autonomous vehicles by overcoming the limitations of single-agent perception. However, most existing approaches assume homogeneous agents, which restricts their applicability in real-world scenarios where vehicles use diverse sensors and perception models. This heterogeneity introduces a feature domain gap that degrades detection performance. Prior works address this issue by retraining entire models/major components, or using feature interpreters for each new agent type, which is computationally expensive, compromises privacy, and may reduce single-agent accuracy. We propose Faster-HEAL, a lightweight and privacy-preserving CP framework that fine-tunes a low-rank visual prompt to align heterogeneous features with a unified feature space while leveraging pyramid fusion for robust feature aggregation. This approach reduces the trainable parameters by 94%, enabling efficient adaptation to new agents without retraining large models. Experiments on the OPV2V-H dataset show that Faster-HEAL improves detection performance by 2% over state-of-the-art methods with significantly lower computational overhead, offering a practical solution for scalable heterogeneous CP.

Figures & tables

Fig. 1: Overview of Faster-HEAL . Stage 1: Homogeneous base training constructs a unified feature space using pyramid fusion and trains the detection head on collaborative data from homogeneous agents. stage 2: For each new heterogeneous agent type, we freeze the pretrained encoder, pyramid fusion extractor/upsampling path, and detection head, and train a lightweight feature aligner, LIFT (visual prompts) to map its intermediate features into the unified space, and the foreground estimator. This design ensures low training cost, preserves the privacy of participating agents by requiring only their intermediate features, and enables scalable heterogeneous collaboration. In our experiments, LiDAR is used in stage 1, while camera or LiDAR could be used in stage 2.

Metric	[email protected](%) $\uparrow$			[email protected](%) $\uparrow$			Model #Training Params (M) $\downarrow$
Based on $m_{1}$ , Add New Agent	$+m_{2}$	$+m_{3}$	$+m_{4}$	$+m_{2}$	$+m_{3}$	$+m_{4}$	$+m_{2}$	$+m_{3}$	$+m_{4}$
No Fusion	74.8	74.8	74.8	60.6	60.6	60.6	/	/	/
Late Fusion	77.5	83.3	83.4	59.9	68.5	68.5	/	/	/
CoBEVT [ 28 ]	82.2	88.5	88.5	67.1	74.2	74.2	33.24	42.05	51.68
HEAL (reported by [ 11 ] )	82.6	89.2	89.4	72.6	81.2	81.3	15.01	1.08	1.91
HEAL (trained with ours configurations)	87.2	91.3	90.9	78.3	84.8	84.5	15.01	1.08	1.91

TABLE I: Detection performance comparison on OPV2V-H using

m_{1}

as the collaborative base and progressively adding three new heterogeneous agents (

m_{2}

m_{3}

m_{4}

). We report [email protected], [email protected], and the model number of trainable parameters for No-Fusion, Late Fusion (no retraining), HEAL [ 11 ] (encoder retraining), and CoBEVT [ 28 ] that retrain their entire model.

Metric	TT $\uparrow$	TF $\uparrow$	PM (GB) $\downarrow$
HEAL [ 11 ]	26.12	13.94	7.0
Faster-HEAL (ours)	30.44	16.63	4.3
Improvement	$\uparrow 1.16\times$	$\uparrow 1.19\times$	( $\downarrow 38.6\%$ )

TABLE II: Training computation cost comparison on OPV2V-H using

m_{1}

as the collaborative base, summed for new heterogeneous agents (

m_{2},m_{3},m_{4}

). TT = training throughput (samples/sec.), TF = TFLOPs (TFLOPs/sec.), and PM = peak memory usage.

Training Components	[email protected](%) $\uparrow$	[email protected](%) $\uparrow$	#TP (M) $\downarrow$
Encoder + BEV (HEAL)	90.9	84.5	18
Encoder + BEV + LIFT	89.6	83.1	18
Encoder + LIFT	92.2	85.3	17.2
BEV + LIFT	90.7	83.6	1.14
LIFT (ours)	92.7	85.7	0.34

TABLE III: Ablation study on stage 2 training components for new agent adaptation (

m_{2}

m_{3}

m_{4}

), with

m_{1}

as the collaborative base. We compare freezing or tuning the encoder, BEV backbone (BEV), and LIFT. The aligner is trained in all cases.

R	[email protected](%) $\uparrow$	[email protected](%) $\uparrow$	#TP (M) $\downarrow$
4	91.6	84.0	0.335
8	92.7	85.7	0.340
16	89.7	82.1	0.352
32	91.4	84.0	0.373
64	90.2	83.0	0.416

TABLE IV: Effect of decomposition rank

R

in PARAFAC decomposition on new-agent adaptation. We report detection accuracy ([email protected], [email protected]) and the number of trainable parameters (#TP). Results show that

R=8

provides the best trade-off The results are reported for

m_{1}+m_{2}+m_{3}+m_{4}

Explore similar work

HeteroPROMPT: A Real-time and Privacy-Preserving Heterogeneous Collaborative Perception Framework
Jul 28, 2026Armin Maleki, Hayder RadhaAutonomous Driving Perception Cooperative Perception
Electrical and Computer Engineering, Michigan State University
GT-Space: Enhancing Heterogeneous Collaborative Perception with Ground Truth Feature Space
Mar 13, 2026Wentao Wang, Haoran Xu, Guang TanHeterogeneous Collaborative Perception Autonomous Driving Perception
Sun Yat-sen University, Shenzhen Campus, Shenzhen, China · Peng Cheng Laboratory (PCL), Shenzhen, China
INTACT: Ego-Guided Typed Sparse Evidence Retrieval for Heterogeneous Collaborative Perception
Jun 3, 2026Chen Li, Shengrong Yuan, Jialong Zuo +3Communication-Efficient Collaborative Perception Cooperative Perception
National Key Laboratory of Multispectral Information Intelligent Processing Technology, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology · Zhejiang Normal University

Jul 28, 2026J/K move · Enter open · S save

HeteroPROMPT: A Real-time and Privacy-Preserving Heterogeneous Collaborative Perception Framework

Armin Maleki, Hayder Radha

Electrical and Computer Engineering, Michigan State University

Autonomous Driving Perception Cooperative Perception Heterogeneous Collaborative Perception

Open paper AskPDF arXiv

Collaborative Perception (CP) improves autonomous systems' awareness of their surroundings by sharing sensor data, intermediate features, and detection results. In real-world deployments, however, collaborating vehicles often use heterogeneous sensors, perception models, datasets, and training domains, creating feature-space shifts that degrade downstream fusion and detection. Existing approaches typically retrain fusion and detection components or introduce modality-specific feature interpreters. These methods scale poorly to newly joining agents and often require access to proprietary metadata, raising privacy concerns. We propose HeteroPROMPT, a real-time and privacy-preserving framework for heterogeneous collaborative perception. HeteroPROMPT rapidly aligns each heterogeneous agent's features with an ego-centric unified feature space through modular prompts and lightweight learning-based tuning, while keeping agent encoders and the collaborative fusion and detection stacks frozen. Its visual prompt-based training and inference modulate Bird's Eye View (BEV) features across channels and spatial locations with low computational overhead. For metadata-free deployment, an autoencoder learns a compact unified representation and extracts modality cues from shared features, enabling real-time modality classification and routing to the appropriate HeteroPROMPT modules without exposing proprietary agent information. Experiments on the OPV2V-H and V2XSet datasets show that HeteroPROMPT improves Average Precision over state-of-the-art heterogeneous CP methods while using orders of magnitude fewer trainable parameters. This offers a scalable and practical CP solution. The proposed modality classifier also predicts the joining agent's modality from compact features with greater than 99.99 percent accuracy during deployment. Code will be available at https://github.com/arminmaleki007/HeteroPROMPT.

Abstract

Explore similar work

HeteroPROMPT: A Real-time and Privacy-Preserving Heterogeneous Collaborative Perception Framework

GT-Space: Enhancing Heterogeneous Collaborative Perception with Ground Truth Feature Space

INTACT: Ego-Guided Typed Sparse Evidence Retrieval for Heterogeneous Collaborative Perception