Driverless vehicles are getting smarter every day, but until mapping and navigational technologies are greatly improved, getting from here to there will still require human intervention for the foreseeable future.
That’s the opinion of John Leonard, roboticist and Professor of Mechanical and Ocean Engineering at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). At a seminar hosted by the MIT Initiative on the Digital Economy last week, Leonard spoke about the current state of navigation and mapping technologies for autonomous vehicles and what’s needed to take them to the next level.
While commercial ventures such as Google’s and Mercedes’ driverless cars, as well as projects by Cadillac and Telsa are highly touted and sophisticated –and some argue that these cars are safer than human drivers — Leonard believes that safety concerns and costs, as well as technology obstacles, will keep the vehicles out of the mainstream for decades to come. “Driving a car or a plane is different than operating an elevator,” he said, and more evidence and discussion are needed before autonomous vehicles proliferate on the nation’s highways and local roads.
“There are too many unexpected problems and dangers,” while driving that require human intervention and judgment, he says. What if there is a construction detour? A speed zone? Bad weather? As for costs, the sensors needed for such precision are very expensive and while they could get cheaper as sales increase, affordability has been an elusive goal to date. “I don’t expect to have truly driverless taxis in Manhattan in my lifetime,” he says.
Lessons Learned From the Darpa Challenge
Leonard clearly believes that self-driving vehicles have the potential to transform our transportation system and he was part of the MIT team participating in the 2006-07 DARPA challenge, that developed and designed a “robocar” Land Rover using a vision-guidance system. It implemented radar and laser scanners to plan and navigate the route, negotiate turns and detect objects on the course. In retrospect, he says, the design was probably too complex and the car finished in fourth place in the competition.
Nevertheless, the experience illustrated the real-world challenges carmakers face in developing viable commercial products. One primary impediment is the need for better computations and understanding of the physical environment and terrain before the cars can make long, complicated journeys.
Autonomous vehicles require computerization of physical, behavioral and procedural actions. The seminar focused on ways to improve navigation and mapping, and techniques for better map-building such as using lidar and vision data in large-scale dynamic environments. In addition, he discussed localization, where the aim is to compute the position of an autonomous vehicle with respect to a previously built map. He also raised some of the open legal, ethical, security, and economic questions associated with self-driving cars, and their potential impact on the labor market.
The bottom line to me was to go ahead and renew my driver’s license; it looks like my Prius won’t be doing errands for me unattended for quite some time. Then again, progress in this field has been exponential. How soon do you expect to see versions of autonomous vehicles on the road?