Stanford's autonomous DeLorean can't time travel, can do donuts
It doesn't have a flux capacitor and may not be able to travel through time like its inspiration in the 1985 feature Back to the Future, but Stanford University's converted DeLorean Multiple Actuator Research Test bed for Yaw (MARTY) can cut some wicked donuts without the aid of a driver. The creation of professor of mechanical engineering Chris Gerdes and his students, the autonomous, electric, drifting automotive research vehicle is part of a student-driven research project into the physical limits of autonomous driving that aims to improve the safe operation of self-driving cars under all conditions.
Self-driving cars are shifting from fiction to fact, but questions remain on how to build one that can operate safely under all conditions. According to Gerdes, the laws of physics place limits on a car's abilities, but with the proper software, autonomous vehicles can push to the edge of those limits to avoid accidents.
Moving incrementally towards these limits is one way to discover how far autonomous cars could go, but Gerdes and his team started at the other end of the scale – they opted to emulate the style of drift racers ... without the driver.
MARTY started out life in 1981 as a standard DeLorean and was converted by Silicon Valley start up Renovo Motors into an advanced rear-wheel drive electric vehicle putting out 4,000 ft-lb of torque. With all of its systems running through a central application programming interface, it's capable of the precise control of forces needed for drift maneuvers. This also made it possible for the Stanford team to integrate autonomous systems in a matter of months, letting the student engineers concentrate on subsystems and algorithms.
Gerdes says the principle behind MARTY is similar to that of a modern car's Electronic stability control (ESC), which monitors the vehicle's handling and controls the brakes and throttle to keep it inside of safe parameters. However, MARTY goes beyond this by embracing instability rather than countering it.
"In our work developing autonomous driving algorithms, we've found that sometimes you need to sacrifice stability to turn sharply and avoid accidents," says Gerdes. "The very best rally car drivers do this all this time, sacrificing stability so they can use all of the car's capabilities to avoid obstacles and negotiate tight turns at speed. Their confidence in their ability to control the car opens up new possibilities for the car's motion. Current control systems designed to assist a human driver, however, don't allow this sort of maneuvering. We think that it is important to open up this design space to develop fully automated cars that are as safe as possible."
For the Stanford team, the tricky bit was getting MARTY to learn when to maintain stability and when to lay it aside. It's this decision making process that the engineers see as key to making self-driving cars safe in all situations.
"When you watch a pro driver drift a car, you think to yourself that this person really knows how to precisely control the path and angle of the car, despite how different it is from normal driving," says Jonathan Goh, a mechanical engineering graduate student in Gerdes' Dynamic Design Lab. "The wheels are pointed to the left even though the car is turning right, and you have to very quickly coordinate the throttle and steering in order to keep the car from spinning out or going the wrong way. Autonomous cars need to learn from this in order to truly be as good as the best drivers out there."
At the moment, MARTY is confined to doing donuts on the track, but Gerdes hopes one day to pit it against a professional driver under actual race conditions, where it will have to deal with obstacles and tight turns at speed. He believes that this would show the car's ability to not only control itself, but also to anticipate how the other driver is acting.
"A drift competition is the perfect blend of our two most important research questions – how to control the car precisely and how to design automated vehicles that interact with humans," says Gerdes. "While we aren't picturing a future where every car produces clouds of white tire smoke during the daily commute, we do want automated vehicles that can decipher the subtle cues drivers give when driving and incorporate this feedback when planning motion. Drifting is a way to study these larger questions, with style."
The video below shows MARTY going for a spin.