Evasive Maneuvers and Drifting for Autonomous Driving

MRSD PROJECT Fall '16/ Spring '17

Abstract

Autonomous vehicles are currently being tested and deployed in major cities around the world, and it is not difficult to imagine that they will be fully deployed on public roads within the next 5-10 years. While these vehicles have advanced sensing capabilities and intelligent systems for decision-making and control, they might still be susceptible to external threats on the road, including reckless human rivers and suddenly-appearing wildlife.

We propose a system that can plan evasive maneuvers through extreme situations such as drifting for autonomous vehicles to counteract these threats. Our system capitalizes on the exceptional sensing and computational capabilities of autonomous vehicles to accurately identify and track any sudden or dynamic
obstacles, and at the same time plan and generate trajectories for the vehicle to evade these obstacles. We demonstrate that our system is able to perform these maneuvers in a swift and precise manner that is unattainable by typical autonomous driving systems or human drivers.

In this project, we implement ed this system on a 1/10 scale RC car with on-board sensors and computation. In our final evaluation, we demonstrated that our robot could avoid a suddenly-appearing obstacle 1.5m away while travelling at 3m/s - a maneuver that is impossible to pull off by automatic braking or human operation.

Go to Summary

Summary

System Summary
Problem Description
Use Cases

Go to Design

Design

System requirements
Functional architecture
Cyberphysical architecture

Go to project management

project management

Schedule
Test Plan
Issues Log
Parts List

Go to documents

documents

Design brainstorming
Drawings, schematics, & datasheets
Experimental Results
Software
Presentations

Project overview

Evasive Maneuvers and Drifting for Autonomous Vehicles

We developed a hardware and software platform that enables autonomous execution of complex maneuvers for RC vehicles. Complex maneuvers may include drifting to evade a suddenly-appearing static obstacle, or parallel parking in a single 180 degree drift turn.

  • A 2D LIDAR provides precise odometry, localization and state estimation for our robot.

  • Autonomous path planning which takes advantage of drifting when presented with obstacles.

  • Detecting and tracking suddenly appearing obstacle in front of the car at high speeds.

  • An iterative Linear Quadratic Regulator generates feasible control sequences while taking non-linear system dynamics into consideration.

Our Team

The roboticists (and gearheads) behind the effort

Amit Bansal

ab1@andrew.cmu.edu

Aum Jadhav

ajadhav@andrew.cmu.edu

Kazuya Otani

kotani@andrew.cmu.edu

Cyrus Liu

xiyuanl1@andrew.cmu.edu

Max Hu

yuanh@andrew.cmu.edu