aiqbench analysis
CoDrive advances generative AI for autonomous driving by solving the critical challenge of maintaining world consistency across multiple vehicles simultaneously, rather than generating isolated ego-vehicle views. By integrating global self-attention mechanisms with precise trajectory control in a shared coordinate system, this framework enables realistic multi-agent simulation essential for training robust cooperative driving policies. The introduction of CoDrive-Bench provides a necessary benchmark to evaluate geometric and instance-level consistency in complex, multi-vehicle scenarios.
CoDrive: Cross-Vehicle World-Consistent Video Generation with Precise Trajectory Control for Cooperative Driving
arXiv:2609.34749v1 Announce Type: cross Abstract: Real-world driving is inherently multi-agent, yet most existing driving world models generate observations from a single ego vehicle. Independently extending them to multiple vehicles does not ensure that different agents observe a consistent shared world. We present CoDrive, a cross-vehicle, multi-view driving video generation framework that jointly generates observations of vehicles sharing the same dynamic scene with precise camera-trajectory control. CoDrive interleaves local self-attention, which models spatiotemporal dependencies among the views of each vehicle, with global self-attention, which enables information exchange and consistency modeling across vehicles. To explicitly encode their spatial relationships, all camera trajectories are represented in a shared world coordinate system and injected into the attention layers through projective relative positional encoding. We further adopt a progressive mixed-task training strategy that combines large-scale real-world single-agent data with synthetic cross-agent interaction data, allowing the model to benefit from real-world appearance distributions while learning cross-agent consistency from simulation. For systematic evaluation, we introduce CoDrive-Bench, a benchmark covering real and synthetic multi-vehicle scenarios and evaluating trajectory controllability, scene geometry consistency, and instance-level consistency. Experiments show that CoDrive improves trajectory controllability and cross-agent geometric and instance consistency while maintaining competitive visual quality.
Abstract:Real-world driving is inherently multi-agent, yet most existing driving world models generate observations from a single ego vehicle. Independently extending them to multiple vehicles does not ensure that different agents observe a consistent shared world. We present CoDrive, a cross-vehicle, multi-view driving video generation framework that jointly generates observations of vehicles sharing the same dynamic scene with precise camera-trajectory control. CoDrive interleaves local self-attention, which models spatiotemporal dependencies among the views of each vehicle, with global self-attention, which enables information exchange and consistency modeling across vehicles. To explicitly encode their spatial relationships, all camera trajectories are represented in a shared world coordinate system and injected into the attention layers through projective relative positional encoding. We further adopt a progressive mixed-task training strategy that combines large-scale real-world single-agent data with synthetic cross-agent interaction data, allowing the model to benefit from real-world appearance distributions while learning cross-agent consistency from simulation. For systematic evaluation, we introduce CoDrive-Bench, a benchmark covering real and synthetic multi-vehicle scenarios and evaluating trajectory controllability, scene geometry consistency, and instance-level consistency. Experiments show that CoDrive improves trajectory controllability and cross-agent geometric and instance consistency while maintaining competitive visual quality.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2609.34749 [cs.CV]
(or arXiv:2609.34749v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.34749 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 290 words
Related coverage
Scale AV Perception Across Vehicle Platforms with NVIDIA Omniverse NuRec
NVIDIA · Aug 31, 2026
VehDyn: A Driving World Model Benchmark for Vehicle Dynamics
arXiv · Sep 29, 2026
NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI
NVIDIA · May 21, 2026
AccidentSim: Generating Vehicle Collision Videos with Physically Realistic Collision Trajectories from Real-World Accident Reports
arXiv · Sep 7, 2026