Skip to content Skip to navigation

Model Predictive Control for Safe Limit Driving in Highly Dynamic Environments

Principal Investigator:

Chris Gerdes

TRI Liaison:

Avinash Balachandran, Carrie Bobier-Tiu & Jon Goh

Project Summary

Leverage Nonlinear Model Predictive Control to safely navigate vehicles through dynamic environments at their handling limits in two different applications: guiding human drivers through obstacle avoidance in urban environments and performing tandem drift maneuvers with a human driver in a head-to-head competition like Formula Drift.

Shared autonomy solutions such as the guardian angel system require tight integration between human intent and machine action, especially at the limits of handling. The MPC framework and open-source results should enable easy translation of research results to TRI.

Research Goals

  • Develop algorithms capable of producing high sideslip angle while path tracking (drifting) and of mirroring a car driving this way while avoiding collision (following).
  • Implement these algorithms on a Toyota testbed with sufficient engine horsepower and demonstrate automated two-car drifting.
  • Adapt prior work with MPC for shared steering control to a nonlinear MPC framework capable of coordinating lateral and longitudinal vehicle motion.
  • Demonstrate driver assistance with this NMPC framework experimentally in a series of scenarios based on real-world collision avoidance, getting progressively closer to the absolute handling limits.