With the era of autonomous cars almost upon us, engineers at Stanford University are already working on something more difficult – robots that can share the pavement with pedestrians. Jackrabbot may look like a backyard BB8 with WALL-E's head stuck on, but its function goes beyond cuteness. It's designed to interact with pedestrians and learn from them how to get around without bumping into people or annoying them.
Given that autonomous cars are giant pieces of metal designed to drive on crowded roads at high speeds, it would seem that making one that could handle a sidewalk to be relatively easy. But the Stanford School of Engineering team says that's not true. In fact, the cars have a lot of advantages because road vehicles run along very carefully laid out routes and they have to follow minutely defined rules. There may be things like dogs leaping into traffic to be accounted for, but they're relatively rare and are more like anomalies than routine problems.
Pedestrian traffic, on the other hand, is an absolute nightmare for robots, even if they are doing something as simple as delivering pizza. How people get about on the pavement is pretty much up to them, with very few explicit rules to follow no matter how crowded the precinct is. Instead, humans rely on all sorts of unwritten rules and unspoken cues that even the walkers themselves might be totally unaware of. Given how few pile ups and punch ups happen on street corners, this seems to work. Most of the time.
Named after the jackrabbits that take up residence on the Stanford campus, Jackrabbot is a 3-foot tall experimental prototype that's engineered to learn the rules of the pavement and to be as friendly as possible with the humans it meets. Its diminutive height, cuddly shape, jaunty hat, and necktie are deliberate features that the team came up with to make it friendly and non-threatening, as demonstrated by the hugs it gets as it roams around campus.
Jackrabbot works by means of a suite of cameras and navigational sensors that allow it to negotiate outdoor pathways, streets, and indoor hallways. Since it is a "social robot," it's programmed to observe human pedestrian etiquette and imitate it. This way, it's hoped that it will eventually learn enough about the unwritten rules of the road to not only avoid obstacles in malls, railway stations, and airports, as well as avoiding hurting anyone's feelings. This may involve everything from learning how to negotiate a ramp, deducing rights of way, waiting its turn at an entrance, or joining a spontaneously begun queue.
The robot does this by collecting images of people as they move about the campus. From these, it maps coordinates and then its algorithms deduce what the proper etiquette is in a particular situation.
The project is still in its early days as Jackrabbot learns how to navigate and builds up its library of social rules. The hope is that such an understanding can later be transferred to other robots, which in five or six years could cost around US$500 for commercial models.
"It's possible to make these robots affordable for on-campus delivery, or for aiding impaired people to navigate in a public space like a train station or for guiding people to find their way through an airport," says Silvio Savarese, assistant professor of computer science and director of the Stanford Computational Vision and Geometry Lab.
The team's research will be presented in a paper at the Computer Vision and Pattern Recognition conference in Las Vegas on June 27.
The video below introduces JackRabbot.
