CooperDrive: Enhancing Driving Decisions Through Cooperative Perception

Deyuan Qu Qi Chen Takayuki Shimizu Onur Altintas

Toyota Motor North America, InfoTech Labs

IEEE International Conference on Robotics and Automation (ICRA) 2026 ITS World Congress 2025 Toyota Web Exhibition

Real-Vehicle Demo

Real-world closed-loop deployment of CooperDrive in occlusion-heavy and non-line-of-sight intersection scenarios. Compared with ego-only perception, cooperative perception enables earlier hazard awareness and safer path planning decisions.

+4.57 s
Minimum TTC gain
89 ms
Average end-to-end latency
~90 kbps
Communication bandwidth
Real-world
Closed-loop validation

Method

CooperDrive method overview

Reconstructed Bird’s-Eye View (BEV) representation generated through cooperative perception, where each vehicle contributes precise localization and detection outputs from our proposed Multi-task BEV Perception Network to enhance situational awareness and support safer path planning.

Highlights

Multi-task BEV perception

A shared BEV backbone jointly supports 3D object detection and localization in one efficient perception pipeline.

Lightweight object-level sharing

CooperDrive exchanges compact object-level results instead of raw sensor data or heavy BEV features, making communication practical for real vehicles.

Planning-aware benefit

Cooperative perception directly improves downstream planning by revealing occluded hazards earlier, without changing the existing planner architecture.

Real-world validation

The framework is validated through real-vehicle closed-loop tests in occlusion-heavy and NLOS intersections, demonstrating practical safety gains under strict communication constraints.

Planning Results

Method TTCmin (s) ↑ DRAC (m/s²) ↓ DCZ (m) ↑ VR (%) ↓
Ego-only 2.05 0.312 2.01 18
CooperDrive 6.62 0.060 4.12 2
Improvement +4.57 -0.252 +2.11 -16

Across all tested scenarios, cooperative perception improves safety margin and enables earlier, less aggressive planning maneuvers.

Metric definitions.

  • Minimum Time-to-Collision (TTCmin): shortest predicted time to a potential collision.
  • Deceleration Rate to Avoid Crash (DRAC): minimum deceleration required to avoid a crash.
  • Distance to Conflict Zone (DCZ): shortest distance to a predicted conflict zone.
  • Violation Rate (VR): percentage of planning steps that violate safety constraints.

BibTeX

If you find CooperDrive useful in your research, please consider citing our paper:

@inproceedings{qu2026cooperdrive,
  title     = {CooperDrive: Enhancing Driving Decisions Through Cooperative Perception},
  author    = {Deyuan Qu and Qi Chen and Takayuki Shimizu and Onur Altintas},
  booktitle = {2026 IEEE International Conference on Robotics and Automation (ICRA)},
  year      = {2026}
}