Deep learning algorithms for driver behavior predictions

This technology consists of algorithms created using physics-based driving models and deep neural networks to predict driver behavior

Unmet Need: Accurate and efficient prediction of driver behavior

Current technologies to predict driver behavior use either physics-based models, which consist of assumptions that over-simplify results, or data-driven models, which require large datasets that can be expensive and time-consuming to acquire. As a result of these shortcomings, there is currently no standard method to accurately, efficiently and cost-effectively predict driver behavior.

The Technology: Physics-informed deep learning algorithm to predict driving behavior

This technology consists of machine learning algorithms that combine physics-based models with historical driving data and deep neural networks. The algorithms are coded in the TensorFlow/Python platform, and can be packaged for use in various programming languages such as R and Python. This allows for integration into other applications to predict driving maneuvers and to identify abnormal driving behavior.

The algorithms have been validated using the publicly available dataset NGSIM.

Applications:

• Predict driving behavior
• Identify drivers with abnormal driving behavior
• Improve road safety
• Use by car manufacturers to monitor driving and improve car safety
• Data for car insurance companies and government agencies
• Predict dementia/neurodegeneration using driving data
• Research tool for studying driving behavior in different situations

Advantages:

• Easily integrated with other platforms or applications that use R and Python
• Accurate and efficient
• Cost-effective
• Scalable and transferrable to other types of datasets

Lead Inventor:

Xuan Di, Ph.D.

Related Publications:

• Mo Z, Shi R, and Di X. “A physics-informed deep learning paradigm for car-following models” Trans Res Part C Emerg Technol. 2021 Sep; 130: 103240.

• Shi R, Mo Z, Huang K, Di X, Du Q. “Physics-informed deep learning for traffic state estimation” Proceedings of the AAAI Conference on Artificial Intelligence. 2021 Jan 17; arXiv: 2101.0658.

Tech Ventures Reference:

• IR CU22032

• Licensing Contact: Dovina Qu