Safe Feedback Interactions in Human-Autonomous Vehicle Systems

Principal Investigators:

Mac Schwager, Mykel Kochenderfer, Marco Pavone, and Chris Gerdes

TRI Liaison:

Project Summary

We seek to develop algorithms for autonomous driving that are provably and practically safe when interacting with other vehicles (both human driven and autonomous) on the road. We consider algorithms that (i) model policies of other drivers, (ii) communicate intent through motion and infer intent of other drivers through motion, and (iii) execute safe actions aware of the dynamic coupling between vehicles.

Providing safe autonomous driving polices that interact intelligently with other drivers on the road will directly enhance Toyota’s autonomous vehicle capabilities, and ultimately lead to a safer driving experience for everyone.

Research Goals

Investigate interactions across a range of timescales
Long Timescales: Interactive Partially Observable
Markov Decision Process (I-POMDP) for Navigation
Intermediate Timescales: Interaction-aware trajectory planning using conditional variational autoencoders and game theoretic planning
Short Timescales: Reactive collision avoidance using buffered Voronoi cells and reachability analysis
Combine algorithms into integrated autonomy stack
Track experiments with full scale autonomous vehicles
Engagement with TRI for tech transfer and infusion