Juntong Peng

Autonomous driving Multi-agent intelligence Foundation model

My Background.

Hi! I am a third-year Ph.D. student in Purdue ECE, co-advised by Prof. Yaobin Chen and Prof. Ziran Wang.

I received my bachelor degree in information engineering from Shanghai Jiao Tong University, advised by Prof. Siheng Chen. I have also spent a wonderful summer at the Department of Computer Science, Purdue University as a research intern, advised by Prof. Aniket Bera.

My research focuses on autonomous driving and computer vision, with an emphasis on multi-agent intelligence, vision-language-action (VLA) models, and foundation models for embodied autonomy. I am broadly interested in building efficient, trustworthy, and human-aligned autonomous systems that can perceive, reason, collaborate, and act in complex real-world environments.

Lately.

Aug 31, 2026	VLM-DC: Vision Language Model for Automotive Data Collection was accepted for publication in the IEEE Internet of Things Journal.
Jul 13, 2026	Four papers were accepted to IROS 2026: FleetAgent, OmniV2X, ViLaD, and Efficient and Explainable End-to-End Autonomous Driving via Masked Vision-Language-Action Diffusion.
Jul 13, 2026	QuantV2X: A Fully Quantized Multi-Agent System for Cooperative Perception was accepted to ECCV 2026.

Perceive. Collaborate. Act.

All publications

IEEE IoT J.

VLM-DC: Vision Language Model for Automotive Data Collection

Can Cui^*, Juntong Peng^*, Sungyeon Park, Yongkang Liu, Seyhan Ucar, Satoshi Kitai, Ahmadreza Moradipari, Maryam Khabazi, and Ziran Wang
^* Equal contribution.

IEEE Internet of Things Journal, 2026

Abs DOI HTML

We introduce VLM-DC, a two-module vision-language model framework for intelligent and targeted automotive data collection. The framework combines cloud-edge multi-agent collaboration with a distributed VLM architecture and knowledge distillation to improve scene selection while reducing storage, model, and bandwidth requirements.
IROS

FleetAgent: Teleoperation Assistant for Autonomous Fleets via Vectorized V2N Messages

Juntong Peng, Qi Chen, Deyuan Qu, Takayuki Shimizu, Yaobin Chen, and Ziran Wang

In 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026

Abs arXiv Code

We propose a cloud-hosted multimodal assistant for autonomous fleet teleoperation that uses compact vectorized V2N messages for efficient narration, explanation, and intervention-priority evaluation.
IROS

OmniV2X: A Generative Foundation Planner for Efficient End-to-End Cooperative Driving

Juntong Peng, Juanwu Lu, Yupeng Zhou, Can Cui, Yaobin Chen, and Ziran Wang

In 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026

Abs arXiv Code

We introduce a generative foundation planner for efficient end-to-end cooperative driving with standard-compliant V2X tokens and lightweight adaptation to cooperative scenarios.
Preprint

Communication-Efficient Multi-Agent 3D Detection via Hybrid Collaboration

Yue Hu, Juntong Peng, Yunqiao Yang, and Siheng Chen

arXiv preprint arXiv:2508.07092, 2025

Abs arXiv

We utilized the confidence and uncertainty of single perception to guide the information sharing process, combining the benefits of both sparse and dense information form. Our method consumes far less communication resources than the previous SOTA.
CVPR

Communication-Efficient Collaborative Perception via Information Filling with Codebook

Yue Hu, Juntong Peng, Sifei Liu, Junhao Ge, Si Liu, and Siheng Chen

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

Abs arXiv Code

We proposed a Communication-Efficient collaboration method, which enhances the tradeoff between performance and communication cost by optimizing information selection and representation.
RA-L

Graph-Based Decentralized Task Allocation for Multi-Robot Target Localization

Juntong Peng, Hrishikesh Viswanath, and Aniket Bera

IEEE Robotics and Automation Letters, 2024

Abs DOI arXiv

We proposed a graph neural network based method to dynamically allocate tasks to different robots. We utilized the comparative advantages of different robots to improve the overall performance of the system.
ICDM

Compatible Transformer for Irregularly Sampled Multivariate Time Series

Yuxi Wei, Juntong Peng, Tong He, Chenxin Xu, Jian Zhang, Shirui Pan, and Siheng Chen

In 2023 IEEE International Conference on Data Mining (ICDM), 2023

Abs DOI arXiv Code

We proposed a transformer-based model to handle irregularly sampled multivariate time series. Our model is inspired by the correlation between multi-agent trajectories.

01 / ABOUTMy Background.

02 / UPDATESLately.

03 / SELECTED WORKPerceive. Collaborate. Act.

My Background.

Lately.

Perceive. Collaborate. Act.