New technique helps autonomous vehicles navigate complex traffic situations

Developed at North Carolina State University, the technique allows autonomous vehicle software to make the relevant calculations more quickly, thereby improving traffic and safety in simulated autonomous vehicle systems.

“Right now, the programs designed to help autonomous vehicles navigate lane changes rely on making problems computationally simple enough to resolve quickly, so the vehicle can operate in real time,” said Ali Hajbabaie, corresponding author of a paper on the work and an assistant professor of civil, construction and environmental engineering at NC State. “However, simplifying the problem too much can actually create a new set of problems, since real world scenarios are rarely simple.”

Hajbabaie continued: “Our approach allows us to embrace the complexity of real-world problems. Rather than focusing on simplifying the problem, we developed a cooperative distributed algorithm. This approach essentially breaks a complex problem down into smaller sub-problems, and sends those to different processors to solve separately. This process, called parallelisation, improves efficiency significantly.”

The researchers have so far tested their approach in simulations, where the sub-problems are shared among different cores in the same computing system. For autonomous vehicles to use the approach on the road, the vehicles would network with each other and share the computing sub-problems.

In proof-of-concept testing, the researchers considered whether their technique allowed autonomous vehicle software to solve merging problems in real time; and how the new ‘cooperative’ approach affected traffic and safety compared to an existing model for navigating autonomous vehicles.

In terms of computation time, the researchers found their approach allowed autonomous vehicles to navigate complex motorway lane merging scenarios in real time in moderate and heavy traffic, with ‘spottier’ performance when traffic volumes got particularly high.

When it came to improving traffic and safety, the new technique is said to have done ‘exceptionally well’. In some scenarios, particularly when traffic volume was lower, the two approaches performed similarly. In most instances, the new approach outperformed the previous model by a considerable margin, the team said. According to NC State, the new technique had zero incidents where vehicles had to come to a stop or where there were ‘near crash conditions.’ The other model’s results included multiple scenarios where there were literally thousands of stoppages and near crash conditions.

“For a proof-of-concept test, we’re very pleased with how this technique has performed,” Hajbabaie said in a statement. “There is room for improvement, but we’re off to a great start.

“The good news is that we’re developing these tools and tackling these problems now, so that we’re in a good position to ensure safe autonomous systems as they are adopted more widely.”

The team’s paper, “Distributed Cooperative Trajectory and Lane changing Optimization of Connected Automated Vehicles: Freeway Segments with Lane Drop,” has been published in Transportation Research Part C.