Automated Co-pilot Can Help You Avoid Collisions
Brett Smith for redOrbit.com – Your Universe Online
With current automobile safety systems like crumple zones and air bags focused on mitigating the damage caused by a crash, MIT researchers have shifted their attention to preventing automobile accidents altogether.
In pursuit of this goal, Sterling Anderson, a PhD student at MIT, and Karl Iagnemma, a principal research scientist in MIT’s Robotic Mobility Group, have developed an automated ‘co-pilot’ that guides the vehicle along a safe route if it detects an impending collision or dangerous maneuver.
The system uses an onboard camera and laser rangefinder to locate potential hazards in the vehicle’s environment. An algorithm then analyzes the data collected by the camera and rangefinder to identify a safe path just ahead of the car. When the driver is about to exit the safety area, the system compensates with the minimal amount of correction necessary to keep the car within the predetermined path.
“When you are driving safely, the system runs in the background monitoring the vehicles environment and your performance,” explained Anderson in an online video. “But if you make a mistake, one serious enough to cause and accident or loss of control, it intervenes to ensure you avoid it.”
Anderson went on to say that the system he is developing is different from the autopilot used by commercial aircraft, which does the majority of the work while the plane is in flight. The aircraft’s system typically hands over control during times of crisis, when a human pilot is expected to react better than the autopilot.
“This abrupt transfer of control, particularly in fast-paced or high-stress scenarios where humans tend to perform poorly, can be tricky,” he said. “Many aircraft crashes in recent memory have been caused by human error in times like these.”
The new MIT system takes the opposite approach, where the human driver is in control by default and the autonomous control only kicks in as needed.
“This allows us to exploit the automated system’s ability to respond quickly and precisely to well-defined control objectives, while exploiting the driver’s uniquely human ability to detect and contextualize patterns and new information, reason inductively, and adapt to new control modes while driving,” Anderson said.
The team has run over 1,200 trials of the system, with few collisions. According to the researchers, most collisions occurred when the vehicle’s camera failed to properly identify an obstacle.
The MIT team also observed an interesting side effect during their experimental trials: Drivers who trusted the system tended to perform better than those who didn’t. For example, when drivers were asked to keep the wheel straight, even in the face of a potential collision, those who trusted the system drove through the course more quickly than those who were skeptical of their robotic co-pilot.
Eventually, the system could become more robust and efficient, researchers said. They noted that it might be possible to trim the system used to identify obstacles down to a single smartphone.
“You could stick your cellphone on the dashboard, and it would use the camera, accelerometers and gyro to provide the feedback needed by the system,” Anderson said. “I think we’ll find better ways of doing it that will be simpler, cheaper and allow more users access to the technology.”
The group presented details of the safety system recently at the Intelligent Vehicles Symposium in Spain.
Image 2 (below): A driver setup tests humans’ response to a new semiautonomous safety system. Credit: Sterling Anderson