At this year’s Autonomous Vehicle Tech Expo Europe, Mircea Gradu, vice president of vehicle engineering at Karma Automotive and chairman of the board at FISITA, will present ‘Advanced connected and automated vehicle (CAV) AI research platform – deployment and results’ as part of the Simulation and Testing, Scenarios & Virtual Validation conference session
Deployed across 25 live public intersections in Orange County, California, the platform combines infrastructure-based lidar sensing, AI-powered traffic intelligence, eco-driving control systems and real-world deployment to create one of the most ambitious connected mobility research initiatives currently operating on public roads.
We caught up with Gradu ahead of the event to learn more about the project and why its findings could have major implications for the future of CAV mobility worldwide.
Could you give us an overview of the CAV AI research platform deployed in Orange County? What problem was it designed to address?
The platform was created to tackle several major urban mobility challenges simultaneously, particularly around traffic inefficiency, vulnerable road user safety, infrastructure intelligence and vehicle energy consumption.
Conventional traffic signal systems rely on static timing plans and outdated loop detector technology, creating unnecessary stopping, queueing and wasted energy at intersections. Meanwhile, pedestrians and cyclists remain vulnerable, with many near-miss incidents not captured. Municipalities lack continuous, real-time, multimodal traffic data needed to optimize signal operations or make informed infrastructure decisions. Vehicle classification, speed, trajectory and behavioral data are often fragmented or unavailable.
Stop-start driving also dramatically increases energy consumption and emissions, particularly for electric vehicles, where efficiency and range remain critical concerns.
What makes this platform unique compared with global CAV initiatives?
There are three core differentiators. First, it operates on public roads, not within a controlled proving ground or isolated test environment. Second is using infrastructure-side lidar sensing rather than vehicle-based intelligence. Third is energy efficiency, mandated through US Department of Energy objectives. Many global testbeds focus purely on autonomy or connectivity. This platform combines safety, sustainability and infrastructure intelligence into one fully integrated deployment.
What were the research objectives, and how do the findings relate to improving safety for cyclists, pedestrians and other vulnerable road users?
One of the project’s most important breakthroughs is shifting safety analysis from reactive to proactive. Historically, safety improvements have relied on analyzing crashes retrospectively, often years later. This platform identifies near-miss events and behavioral risk patterns in real time, fundamentally changing how intersections can be monitored and improved. Crucially, intelligence sits within the infrastructure rather than inside the vehicle. That means road users benefit equally.
This infrastructure-centered approach delivers a level of equitable protection that vehicle-only systems cannot achieve alone.
You’ve analyzed mixed traffic, vehicle speeds and non-university road users. What early insights or behavioral patterns have emerged from the data?
For the first time, pass-through traffic across the University of California, Irvine network can be accurately quantified, giving transportation planners robust evidence to justify signal timing and infrastructure changes.
The platform demonstrates that energy savings and traffic flow improvements are real, although dependent on traffic density. Strongest performance gains appear in low to moderate traffic conditions, highlighting an important challenge for scaling during peak congestion.
Another major shift is moving from crash-based safety analysis toward real-time near-miss intelligence. Rather than waiting years for accident data, transportation teams can identify high-risk behaviors instantly and map them with precision.
Interestingly, the findings reveal that energy efficiency and vulnerable road user safety reinforce one another. Smoother traffic flow reduces vehicle energy consumption and stop-start conditions associated with pedestrian and cyclist conflicts.
Privacy-preserving lidar sensing avoids much regulatory friction faced by camera-heavy monitoring systems, making continuous public road deployment politically viable in many Western cities.
What is the analysis framework behind the project? How do microsimulation, model-based engineering and X-in-the-loop testing contribute to it?
The project combines simulation, model-based engineering and physical vehicle validation into one tightly integrated development framework. One standout result was that the combined vehicle and signal controller delivered a 44% improvement in average vehicle speed while producing the largest energy savings. The correlation between simulation and X-in-the-loop testing results was also remarkably strong, validating the real-world applicability of the modeling approach.
Performance proved density dependent, with the most significant gains achieved under lighter traffic conditions. Sensor capability was also critical. Greater lidar detection distance improved controller performance, strengthening the case for next-generation sensing technologies in future connected infrastructure systems.
How do you balance simulation and virtual validation with real-world scenario testing, and what advantages does this hybrid approach provide?
The platform follows a structured progression from simulation to X-in-the-loop validation, finally to live road deployment. The first phase uses simulation to model eco-driving behavior at signalized intersections under varying traffic conditions and sensing ranges. The second phase introduces a real Scion iQ battery-electric vehicle at Argonne National Laboratory, allowing researchers to measure energy behavior while simulating surrounding traffic. The final phase moves into live deployment across 25 public intersections with real traffic, signal hardware and vulnerable road user interactions. Rather than treating simulation and real-world testing as separate workstreams, each phase directly informs and validates the next.
The agreement between simulation and physical testing was especially impressive. For example, simulation predicted a 31.6% reduction in fuel consumption for the vehicle controller alone, while X-in-the-loop testing measured 26.4%. For the combined vehicle and signal controller, simulation predicted 38.4% savings and testing confirmed 36.1%.
That consistency strongly suggests the simulation models accurately capture real-world vehicle dynamics.
Why did you choose to present these findings at Autonomous Vehicle Tech Expo Europe?
After introducing the project at the event in 2022, we felt this year presented the ideal opportunity to share a comprehensive update on the platform’s deployment, validation and real-world results. The conference consistently attracts an exceptionally knowledgeable audience and creates the perfect environment for deeper technical discussions around connected and autonomous vehicle deployment.
Click here to see the full conference program
Autonomous Vehicle Tech Expo will take place at the Stuttgart Messe in Germany, June 23-25, 2026. Click here to register for your FREE expo pass and click here to purchase a conference pass.
