Random Walk
Automata in Our Midst
Pity the poor service robot. It has its marching orders; it knows where to go, but to get there, it must thread its way through a mob of humans. We all face that same problem daily, in malls and museums, on boardwalks and boulevards: navigating a living, jostling obstacle course. Only this purposeful pedestrian is electronic.
“I see crowds as swarms of massive particles with random trajectories,” explains Caltech graduate student in control and dynamical systems Pete Trautman. “Picture a hapless robot wading into the melee. It takes baby steps. It backpedals frantically.” And one stubborn human, motivated by obstinacy, curiosity, or simple obtuseness, can detain it indefinitely.
Not that most of us flesh-and-blooders wish our mechanical friends ill. Adults frequently try to engage them in conversation, no doubt having been conned into thinking they are sentient by the melodious bleeps and burbles of R2D2 and WALL•E. And University of Washington researchers have observed infants responding to social cues from robots. But crowds are chaotic, and even a mind-reading android capable of predicting each person’s every move might never find a clear path. All the obvious strategies have limitations. “Make an aggressive beeline? That’s obnoxious,” Trautman cautions, “and potentially dangerous. Wide, evasive detours? Inefficient or impossible.” What about simply waiting for the seas to part, then making a dash for it? “And if they never part?” he shrugs. “You’d wait forever. We call that the Freezing Robot Problem.”
Yet humans navigate crowds routinely—how? “First,” Trautman says, “I collected mountains of data by filming pedestrians negotiating a crowded sidewalk.” Their motions revealed details of the hidden logic known as cooperative collision avoidance: the subconscious twists and turns we all execute to keep from continually bumping into one another. Next, he constructed a mini-robot of intentionally disarming cuteness (a laptop computer on wheels, with stereo camera “eyes”) and turned it loose among the lunchtime crowds in Caltech’s Chandler Dining Hall. Some days he hovered nearby, maneuvering it remotely; other times, the WALL•E wannabe moved autonomously. Hearteningly, the seething mass of humanity accepted the newcomer. Far from running interference, they simply flowed around it as if it were any other (very short) chow seeker on a mission. Some even professed not to have noticed it.
These observations led Trautman to a key insight. “We all expect cooperative collision avoidance from robots,” he says. “It turns out they can also expect it from us.” People are surprisingly predictable creatures, he explains. “We want to get where we’re going, we can only change direction so fast, we don’t like touching strangers. Some metal gizmo whizzes up. Maybe we noticed someone with a joystick nearby—whatever. We’re busy, we’re distracted, we make way and move on.” The only exception is when the robot dawdles or appears lost. “Then folks notice it and crowd around. Certain individuals even try to prevent it from moving off again.”
Trautman presented his crowd-navigating algorithm at January’s TEDxCaltech conference. Using probability theory to predict the behavior of others, it charts assertive courses. “In fact,” he notes, “the algorithm generally selects shorter, smoother paths than most humans do.” So next time you hear “Last orders!” or “Look, it’s Brad Pitt!” or “70 percent off in housewares!” . . . follow the nearest robot. —DZ
In this overhead view of part of Chandler’s food-service area, a small crowd gathers around Trautman’s robot (green circle) the moment it stops moving. For videos, see http://www.cds.caltech.edu/~trautman.
