Robot Co-Driver Takes the Wheel to Avoid Crashes
This setup helped researchers test their new semi-autonomous car system in a remote-controlled car.
CREDIT: Sterling Anderson
A future full of self-driving cars may be awesome, terrifying or sad — it depends on whom you ask. But one group of researchers doesn't want this to be an all-or-nothing choice: They are racing to develop the first robotic "co-driver" that works with human motorists. The robot takes over when it senses the car will hit something, saving the motorist from an accident, but otherwise allowing the human driver to make the decisions. Yet at the push of a button, the car is also able to drive itself, much like the Google car project.
"It's capable of not just semi-autonomy and not just full autonomy. It's capable of either," said Sterling Anderson, a doctoral student in mechanical engineering at MIT who created the system.
The benefit of the novel semi-autonomous mode is that it might show up in the market a bit earlier than a wholly self-driving car, Sterling thinks. The semi-autonomous system would be a major safety feature that requires fewer radical legal changes than a fully autonomous car does, he said. [5 Ways Self-Driving Cars Will Make You Love Commuting]
Sterling's system uses a camera, a laser and other sensors to detect lane markings and obstacles in the road. He worked with Karl Iagnemma, an MIT robotics researcher, to write a computer program that uses data from those sensors to figure out a safe zone for the car as it moves.
In semi-autonomous mode, the computerized system takes the wheel only when the car is about to leave the safe zone. If, for example, a deer were to jump in the road suddenly, the system would kick in and steer the car toward safety. Once the danger was over, the car would relinquish control to the human driver again.
Anderson has tested his system about 1,400 times during which the driver operated a car remotely, and 800 times with drivers actually in the car. His testers have included both professional drivers at the Ford Motor Co., which has licensed MIT's patent for the technology, and average people. "Responses have sort of run the gamut," he said.
The system performed well for those who trusted it. For those who were distrustful, the ride was jerky, as they might be trying to steer the wheel after the car had automatically guided them around a barrel set in the testing field, Anderson said.
Still others were lazy. "Like, 'Why even bother driving around this barrel?'" Anderson explained. "That's one of the problems that we're concerned about. We don't want to create lazy drivers or unskilled drivers." He is still working on ways to tailor the system for different drivers' preferences and for drivers who are still learning, who would benefit without computer help.
Though Ford has licensed the patent on this technology, the company doesn't have to tell Anderson and his colleagues when or how it will use its license, so Anderson didn't know exactly when we may see his robotic co-driver on the road. He guesses it'll first show up as a luxury safety feature in about five or 10 years.
A driverless car may render the semi-autonomous portion of his system obsolete in the future, although Anderson doesn't think that will happen for some time. Meanwhile, his automatic co-driver may act as a "bridge to a fully automated future," he said. "It can take us along that road, getting us closer and closer to full automation."