Behavior prediction

In order to operate efficiently while ensuring safety, an autonomous vehicle needs to predict the behavior of other agents in its environment. These agents include human-driven vehicles, pedestrians, bicyclists, etc. As humans, while driving, we instinctively predict other road users’ behavior. But it’s not a trivial task from the perspective of an autonomous vehicle.

We are investigating methods of behavior modeling and motion prediction of road agents. We are looking into different aspects of the problem, including the agent types of pedestrians, vehicles, bicycles, and at many types of traffic scenarios such as roundabouts, junctions, crosswalks, etc. We investigate both the use of infrastructural as well as in-situ sensors for the purpose of agent motion prediction. Such techniques are paramount for pushing autonomous vehicles towards level 5 autonomy, as well as having application for automated vehicles operating at ports, warehouses, factory floors, etc.

We are also investigating many other aspects of autonomous driving, including GNSS-free localization for level 5 autonomy, and novel perception modalities in autonomous driving, such as air flow, etc.

Naveed Muhammad

Associate Professor of Autonomous Driving, Behavior Prediction Team Lead