Robots may be the solution to New York City’s trash problem. In fact, Cornell Tech Ph.D. candidate Frank Bu and his team engineered trash barrel robots — human controlled trash cans that pick up garbage — in New York City this summer in order to study how people interact with robots in public spaces and how this can be used to encourage trash pick up.
He first worked on this project 11 years ago with his faculty advisor at Stanford University where he engineered a trash can on a rumba. After noticing a dire need for trash pickup in the city, Bu and his team contacted Village Alliance, a leading advocate for the Greenwich Village community, in Astor Place, NYC to launch this machine in a public space.
It’s no secret that NYC has a waste problem. Streets can often be seen lined with trash bags full to the brim. According to the Mayor’s Office of Sustainability, the metropolitan region produces 14 million tons every year, much of which lands in landfills or incinerators, or pollutes sidewalks and waterways.
To tackle the issue on a larger scale, Bu repurposed used hoverboards, in lieu of the rumba, and created two different garbage cans instead of just one. “[Hoverboards] are more powerful and faster so you can give the robot more expression of capability,” Bu said. “It can move faster and it can wiggle. Also it carries more weight since hoverboards are meant to carry adults at ten kilometers per hour.”
Bu also mentioned that using hoverboards was a more cost effective method because it was cheaper than buying the two motors and battery separately.
Bu and his team of three undergraduates required two wizards and an interview conductor.
The two “wizards” — engineering student JiaYing Li ’25 and Cornell Tech Human Robotics Interaction Researcher Melina Tsai — controlled the robot from the back end. “Since I’m a human, I know where to not go out of limits,” Li said. “If there were people waving at the trash can, I would come over to them and they would throw the trash.”
Afterwards, Nicole Sin ’26 conducted interviews with users of the robot to learn about their opinions on the trash bot. “Their reaction to [the bot] was mostly positive and they wanted to see more of it. I would ask them if there’s anything they would change and they said ‘we need more of this and this is very convenient for people with disabilities,’” Sin said. “Another point that they really liked was that the trash can stayed within a radius of the park. It wouldn’t go out to the streets and it would just be confined to one space. They felt safe with it.”
The researchers then analyzed the videos to look for covert and overt human gestures used to signal the bot. “There would be strikingly positive interactions where people would motion to the trash can or wiggle their trash in front of it and our wizards would drive it over,” Sin said. “But some of them were more subtle in that they just finished eating and you just had to pick up on it.”
The team also observed different reactions to the trashbot in Astor Place, Manhattan than in Albee Square, Brooklyn, attributing it to different borough demographics.
“In Astor Place, it’s mostly tourists — NYU and Cooper Union [students]. The demographics are definitely different from Albee Square, which has more local people who feel more ownership to the place,” Bu said. “They’re like it’s a New York thing — this is a New York City rumba.”
Some bystanders even thought that the bot was an art piece, mistaking it for the Astor Cube, a prominent sculpture in the Manhattan neighborhood.
On the other hand, Bu said that Albee Square residents have found a historical significance with the bot. He recalled an interview with a truck driver who talked about his previous job cleaning the streets of Albee Square a decade ago. “Ten years later, I still hang out here and it’s nice to see this development happening through time.”
Bu and his team are using these interviews and videos to program different cues into the robot so that the robot can act in socially acceptable ways in public spaces without the need of a wizard. “In the future, we can have a more self-exercising model to decide when to provide the service or come close to a human based on how long they’ve been here, are they eating, do they have trash,” Bu said.