Vehicle Trajectory Prediction is a Critical Safety Component

Partner

USDOT Center for Connected and Automated Transportation

Industry

Transportation

Objective

Improve safety of mixed traffic flows by accurately predicting human-driven vehicle trajectories.

Solution

Predictive Machine Learning Models, Long Short-Term Memory Model, Next Generation Simulation

Focus Area

Machine Learning, Autonomous Vehicle

Download PDF
Read More

The Opportunity

The advent of connected and autonomous vehicles (CAVs) will change driving behavior and the travel environment, providing opportunities for safer, smoother, and smarter road transportation.

The Challenge

According to the US Department of Transportation, in 2016 motor vehicle-related crashes on U.S. highways claimed 37,461 lives. Their research shows that 94% of serious crashes are due to human error, so predicting human-driven vehicle (HDV) trajectories are critical to safety improvements as both HDVs and CAVs will share the road for decades to come.

Predicting the trajectories of surrounding HDVs is critical for CAV control.

“From reducing crash-related deaths and injuries, to improving access to transportation, to reducing traffic congestion and vehicle emissions, automated vehicles hold significant potential to increase productivity and improve the quality of life for millions of people.”

Secretary Elaine L. Chao – U.S. Department of Transportation

Why RDSC?

CAV trajectory data can be leveraged with advances in computing and machine learning algorithms to potentially predict trajectory data of HDVs along multiple parameters, like acceleration and speed. Complex data modeling based on real time inputs requires multiple state-of-the-art machine learning algorithms.

Data Science in Action

Providing autonomous vehicles the ability to communicate over short distances eliminates or reduces many of the drawbacks imposed by “line of sight” technologies and limitations in sensing range. Using a Long Short-Term Memory (LSTM) model, with multiple sensors and vehicle-to-vehicle communications, a CAV can track the trajectories of other CAVs in communication range. Based on these enhanced and improved predictions, CAVs can react accordingly to avoid or mitigate traffic flow oscillations and accidents.

The Result

As a case study, the LSTM model is built based on Next Generation Simulation (NGSIM) dataset. A market penetration rate of 50% is assumed for CAVs. The accuracy of the 10-timestep ahead HDV trajectory prediction can be as good as 95.40%.

What’s Next?

Autonomous vehicle technology has made huge advances in the last few years. However, there are still a lot of challenges in detection, tracking and trajectory prediction of surrounding objects. RDSC scientists will continue to develop advanced computer vision technologies for this task. Reinforcement learning and imitation learning models will then be applied for optimal trajectory planning of autonomous vehicles.