Foundation Models for V2X-based Cooperative Autonomous Driving

Introduction

The 4th edition of the DriveX Workshop focuses on how foundation models and V2X-based cooperative systems can redefine perception, prediction, planning, and decision-making for autonomous driving and intelligent transportation infrastructure.

Traditional single-vehicle pipelines have achieved impressive progress in 3D detection and tracking, yet they remain constrained by limited viewpoints, occlusions, and domain shifts. Cooperative driving systems, powered by V2X communication and roadside/edge intelligence, extend sensing range, enrich scene context, and enable shared representations across vehicles and infrastructure.

In parallel, foundation models, including vision, vision-language, and multi-modal large models, unlock powerful generalization capabilities: open-vocabulary understanding, scalable pretraining, zero-shot adaptation, and interpretable reasoning about complex road scenes. Emerging end-to-end and agentic systems such as large driving models promise unified perception-to-control frameworks but raise new questions in trustworthiness, reliability, calibration, and evaluation at urban scale.

DriveX 2026 convenes researchers and practitioners from computer vision, robotics, communications, transportation, AI safety, and policy to:

• Design foundation models natively aware of cooperative perception and real-world V2X constraints.
• Bridge large-scale datasets, simulation, and deployment for multi-agent coordination.
• Discuss standards, benchmarks, and open challenges for safe, human-centric cooperative autonomy.

Topics of Interest

• Foundation models for cooperative autonomous driving and intelligent transportation systems
• Vision-language models for traffic scene understanding and explanation
• LLM- and agent-based support for perception, prediction, planning, and V2X coordination
• Cooperative perception, V2X communication, and infrastructure-assisted driving
• Datasets, benchmarks, and evaluation protocols for foundation models and cooperative perception
• 3D occupancy, 3D detection, 3D semantic segmentation, and holistic 3D scene understanding
• End-to-end pipelines and large driving models for multi-agent decision-making
• Safety, robustness, uncertainty estimation, and open-set reasoning in cooperative systems
• Vehicle-to-Infrastructure (V2I) and Vehicle-to-Everything (V2X) interaction and standards

Schedule (Tentative)

Final schedule, room allocation, and speaker order will be announced closer to the workshop date.

Confirmed Keynote Speakers

Paper Track

DriveX 2026 invites high-quality contributions on foundation models, V2X-based cooperative perception, large driving models, and related topics outlined above.

We welcome:

• Novel full papers (max 8 pages, excluding references) for publication in the official proceedings.
• 4-page extended abstracts or 8-page versions of recently published work (non-archival, not included in proceedings).

Submissions must follow the official CVPR 2026 style: LaTeX or Typst.

• Submission Portal: tbd
• Submission Opens: February 1, 2026 (23:59 PST)
• Paper Submission Deadline: March 1, 2026 (23:59 PST)
• Notification to Authors: March 10, 2026
• Camera-Ready Deadline: March 20, 2026

Paper Awards

• 🏆 Best Paper Award
• 🏆 Best Paper Runner-Up Award
• 🏆 Best Poster Award
• 🏆 Best Keynote Presentation Award

DriveX Challenge

The DriveX Challenge fosters rigorous, reproducible benchmarking of cooperative perception and planning on real-world V2X datasets. Tracks are designed in close collaboration with dataset creators and industry partners.

• V2I-Based Cooperative Perception
  Infrastructure–vehicle fusion using TUMTraf V2X CP. Focus on cooperative 3D detection and tracking with infrastructure-mounted LiDAR, radar, and cameras, emphasizing occlusion handling, long-range awareness, and reliability under real-world conditions.

• Accident Scene Understanding & Safety Reasoning
  Built upon TUMTraf Accid3nD. Participants design models for high-risk scenarios, proactive risk assessment, and early accident prediction using cooperative perception signals to support Vision-Zero mobility.

• End-to-End Multi-Agent Autonomous Driving
  Using V2XPnP and V2V4Real, teams explore end-to-end policies and trajectory planning with single-vehicle, multi-vehicle, and vehicle–infrastructure inputs. The track highlights how cooperative intelligence improves policy learning, coordination, and safety.

Competition Timeline

• Competition Announcement: January 1, 2026
• Submission Deadline: March 1, 2026 (23:59 PST)
• Notification to Participants: March 31, 2026

Top-performing teams will be invited to present at the workshop. Detailed rules, baselines, and submission instructions will be released on the official challenge page.

Challenge Awards

• 🏆 Challenge Winner Award (1st Place): Certificate, recognition, and prize

Organizers

Invited Program Committee

Sponsors

DriveX 2026 welcomes sponsorship from industry, startups, and institutions interested in foundation models, cooperative perception, simulation, and large-scale autonomous driving systems.

For sponsorship opportunities, please contact: walter.zimmer@cs.tum.edu.