Mild-Mannered Robot Studies Pedestrian Behavior

Friendly-looking robot is using machine learning algorithms to learn how to move easily among humans.

With expressive eyes, a squat body and a two-fingered arm, the JackRabbot 2 robot is learning how to share the sidewalk with other pedestrians. It is the latest tool in a field of research focused on creating generalized personal helpers for humans that will deliver packages, do household chores and perform simple day-to-day errands. To perform these tasks, these robots will have to learn to work close to, and around, people.

“The JackRabbot project is developing a robot that doesn’t just navigate an environment by following the behavior of a traditional robotic system, such as going from Point A to Point B and avoiding bumping into obstacles,” said project leader Silvio Savarese, associate professor of computer science at Stanford University. “We want a robot that is also aware of the surroundings and the social aspects of human-robotics interactions, so it can move among humans in a more natural way.”

The JackRabbot 2 uses algorithms based on machine learning techniques that will allow it to navigate the unwritten rules of pedestrian traffic—social conventions that are far more complex than lane markings and traffic lights.

The researchers estimate they will need at least 24 hours of data to teach the algorithm that will allow the robot to navigate crowds with humanlike etiquette. The data includes pre-existing videos featuring different modes of transportation: walkers, skateboarders, bicyclists and people on scooters. The developers will also use data from a video game in which people maneuver a simulation of JackRabbot 2 through online environments. They will also take the robot out for walks, navigating it manually in the way they want it to someday move by itself.

The robot features an arm and a face, an upgrade from the original JackRabbot. The robot’s arm, usually used to interact with physical objects such as opening a door, will also be used to convey intent—signaling a person to go ahead or stop with hand gestures. JackRabbot 2 will also use facial expressions and sounds to communicate with people nearby.

JackRabbot 2 is equipped with multiple sensors to help it navigate its environment, including depth-sensing and stereo cameras, GPS and three LiDAR sensors.

The original JackRabbot, the prototype for a new generation of “social robot” designed to learn how to move among humans.

The most immediate application of the JackRabbot project is the “last-mile problem”—bridging the gap between large-scale delivery hubs and individual businesses or homes. A robot like JackRabbot’s may someday be taking packages from an autonomous delivery truck to your front door. The researchers are hoping to test this capability by using JackRabbot 2 to deliver packages on campus.

Eventually, one of JackRabbot 2’s descendants may well be your butler.

Read more about people-friendly robots at Vector Wants to Be the Home Robot that Feels Alive.