While Google’s bubble-shaped car putters around in California and Tesla’s Autopilot guides cars into garages, Georgia Tech is taking autonomous driving to the extreme. A team of researchers from both the Daniel Guggenheim School of Aerospace Engineering and the School of Interactive Computing at the university have built AutoRally, an autonomous rally truck that pushes the limits of its performance.
Okay, it’s a one-fifth scale rally truck robot weighing about 48 pounds built on an RC chassis. But the three-foot-long truck can drift and jump just like a full-scale rally truck could. And just like a real rally car, sometimes it hits barriers and misjudges curves.
The pair of AutoRally vehicles built by the team race around the test track at the Georgia Tech Autonomous Racing facility at a scale equivalent of 90 mph (actually 20 mph), so the aluminum body of the vehicles is rather tank-like. It has to protect the processor, battery, GPS, cameras, and sensors that make up the brains of the robots, which calculates about 2000 possibilities every 50 milliseconds.
In addition to just being cool, these AutoRally robots serve a research purpose. The Google and Tesla cars mentioned above are operating in very controlled environments while the autonomous technology on board is developed. But fully autonomous cars are going to be required to handle very complicated driving tasks, and practicing with a one-fifth-scale robot is safer than trying to push the performance envelope with a human in the vehicle.
The goal of the researchers was to develop algorithms that could handle the complicated and quick decision-making process that takes place in a human brain when driving gets tricky. Autonomous systems can handle the easy stuff, like driving at a steady speed on the highway or creeping along in gridlock. But when a car needs to take an evasive maneuver at speed—to avoid an oncoming car or swerve to miss a deer, for example—current autonomous systems need the human driver to take over.
What better way to develop these high-performance algorithms than with a robot rally car? The AutoRally cars use handling data and the dynamics of the vehicle to determine the smoothest, most stable path around the track, even at speed around corners. Sometimes drifting around a dirt corner is the best—and coolest—choice.
Want to build your own AutoRally vehicle for research and testing? Specs and code are up on Github.