Autonomous Drifting: Demonstrating a Reciprocal Architecture for Motion Planning and Control

Stanford Investigators 

• Chris Gerdes
• Grace Gao

TRI Investigators

• Avinash Balachandran
• Carrie Bobier-tiu
• Jonathan Goh
• Matthew Brown

Project Summary

Automated vehicles, whether guided by a human driver or operating autonomously, must move safely and effectively through a highly uncertain environment. This motion planning problem, however, possesses a very special structure. Dynamics that follow Newton’s second law with

certainty separate the uncertainty about the vehicle’s location and the locations and actions of other actors in the environment from the uncertainty of the available tire forces and the actuator inputs needed to produce those forces. This structure suggests an architecture that reciprocally couples these layers to provide modularity and full performance up to the vehicle handling limits.

This project seeks to develop an architecture for automated vehicle motion planning and control that accounts for uncertainty at the proper level and rigorously handles the interactions between planning and safe operation up to the tire friction limits. The envisioned architecture consists of a vehicle-independent planning layer that receives knowledge of the achievable acceleration and jerk and uses that to plan a trajectory through an uncertain environment. This layer couples reciprocally with a vehicle dynamics controller that achieves the desired acceleration and jerk while communicating back to the planning layer the knowledge of achievable input it requires.

The project involves Stanford and TRI working together on a common Supra development platform to realize this vision.