The Navlab 11 vehicle
Many inventions we now take for granted, such as the typewriter and speech-to-text technology, were developed as assistive devices for the handicapped before revolutionizing the daily lives of everyone else.
The same may be true for a new development from Arne Suppé, Luis E. Navarro-Serment, and Aaron Steinfeld of the Robotics Institute at Carnegie Mellon University. The researchers have successfully tested a system that allows a car to park itself using remote video supervision via an iPod Touch. The purpose of the device isn’t convenience, but enablement for handicapped individuals. (The number of Americans in wheelchairs is growing faster than the population, and even faster than the accelerated population growth of elderly baby boomers.) Handicapped individuals typically do not have the room to get a wheelchair out of a vehicle parked in a cramped parking lot. The car could drop off the user curbside and then park itself. Retrieval of a vehicle would be similarly automatic.
In the mid 1990’s this group of researchers also invented a system for automatic parallel parking that was deployed commercially a decade later.
In order to accomplish the new feat, the researchers used the Navlab 11 vehicle, a Jeep Wrangler tricked out with an array of sensory and processing capabilities, including “six SICK LMS 2-D laser scanners, a custom vehicle state measurement system based around GPS, inertial, and odometry sensors, and various cameras, all mounted on a reconfigurable sensor rack. A small array of on-board computers record and analyze sensor data.”
“Multiple behaviors are used in a coordinated fashion to drop off a passenger, park the vehicle, and return for passenger pick up.”
While the vehicle is fully capable of parking itself with no human involvement at all, the researchers designed the system so that a human could both watch the parking process from afar and interrupt it if necessary. Previous research suggests that people won’t accept a fully autonomous technology, because of fears about loss of control and legal liability, and, the researchers admit, even a vehicle as well-equipped as their demonstration platform sometimes makes mistakes.
Desire for override authority notwithstanding, current technology is not ready to exclude the human from the control loop. Perception systems are still error-prone. For example SICK LIDARs often have trouble detecting black vehicles that have extremely low albedo in the infrared. Of greater concern is the general difficulty of autonomy in unstructured situations, which is sometimes the case in parking lots, as opposed to highways where the behavior of vehicles are much more predictable. At this time, keeping the human in the loop is a requirement.
An iPod touch interface “shows 60-degrees of forward view at a time and can be panned left and right to reveal a full 180- degree [field of view] image.”
The synthesis of the vehicle and remote operation via an iPod touch is a system that doesn’t require a human to “drive” the vehicle remotely–which would be challenging given the limited situational awareness afforded by a video feed–but allows the human to intercede should the situation become more complicated than the vehicle can handle on its own. (For instance, if a pedestrian or another car get in the way of the parking vehicle.)
Swiping up or down [on the iPod remote video feed] intuitively toggles between forward and back video, echoing the convention of checking a rear-view mirror. Video is transmitted via WiFi datalink, however in practice, we expect the datalink will be whatever is most suitable for the user (e.g., 4G wireless, DSCR, etc).
Here’s a video of the system in action. (The experimenter in the driver’s seat is there only for safety reasons; to take over if the system fails.)
Future work, say the researchers, will revolve not around improving the system, which they argue is quite robust, but conducting research on potential users in order to figure out how to overcome their apprehension about adopting such a system. If the group’s past success with parallel parking is any guide, role out of a commercial vehicle could arrive in as little as a decade.
Original paper: Semi-Autonomous Virtual Valet Parking, Arne Suppé, Luis E. Navarro-Serment and Aaron Steinfeld, of the Robotics Institute at Carnegie Mellon University