Tesla has their Autopilot to guide cars into garages and Google has those bubble-shaped cars. But when it comes to autonomous cars, Georgia Tech is taking things to the extreme with their autonomous rally truck, the AutoRally.
Granted, the AutoRally is not a full-scale rally car; it’s only one-fifth the size. However, it scales accurately. The team raced it at 20mph, which equates to 90mph for an actual car. Its sturdy aluminum body protects the cameras, GPS, sensors, battery and processor that constitute the robot’s brain, which calculates roughly 2000 possibilities for every 50 milliseconds.
This lets the car react to unexpected situations and preserve control when it is close to its friction limits. Its components also remain safe if it crashes; even if the crash is nasty.
Other than having the world’s coolest autonomous car, Georgia Tech built their two AutoRally cars to handle the quick and complicated decision-making process that humans do when faced with challenging situations as they drive.
To achieve this, they used a system called Model Predictive Path Integral Control (MPPI). This calculates the most efficient and stable way to handle a number of scenarios. It allows the car to handle all of its computational decisions in real time.
The AutoRally car’s initial training phase comprises of a regular person–not a professional rally driver–driving it around the track with a remote control. Despite this, the car managed to develop abilities like power sliding on its own. The car did crash a few times.
However, this was mostly because of software crashes and the car having trouble adapting to a changing track surface. The developers made some changes to the software and now the AutoRally can handle a track with unpredictable terrain changes.
If you wish you had your own autonomous rally truck, you can download the software from their website and follow the tips.
Check out this quick video: