Predictable and Customizable Autonomous Driving

About this Project

At a glance

As autonomous cars become better and better at planning trajectories that reach the destination and avoid collisions, the time will quickly come to look beyond the function of these trajectories, to what they implicitly communicate to humans: to the passengers, to pedestrians, to other drivers. All these people are trying to reason about the car’s actions. When an autonomous car at a stop sign inches forward, it does not just achieve the functional goal of changing its location closer to the destination: it also communicates to pedestrians that it will go next instead of allowing them to cross. When it accelerates, it not only moves faster, but it communicates to its passengers that it is confident in its plan. When it oscillates between accelerating and breaking, it conveys hesitation, which can be very important in prompting other cars and pedestrians. Whether we plan for it or not, the car’s motion will communicate to people. The only question is, are we going to enable cars to take control of this communication or not? This project proposes that we do. This research will go beyond functionality, and weave interaction with people into the motion planner itself.

Research on interaction and communication through motion has either focused on autonomously conveying the goal state of a motion early on, or on studying how a few handcrafted features influence perception of certain emotions. Here, the team will take a learning trajectory optimization based approach to generate expressive motion based on crowd-sourced data.
Principal InvestigatorsResearchersThemes
Anca DraganChandrayee Basu
Dorsa Sadigh
Sanjit Seshia
Autonomous Vehicles