Technology

Meet ‘DribbleBot,’ MIT’s soccer-playing robot

Lessons learned creating a robot that can move a ball through mud and snow could help scientists create life-saving robots in the future.

Mike Grimmett/MIT CSAIL

April 3, 2023

Scientists at MIT have created a robot capable of dribbling and running with a soccer ball that could pave the way for the creation of life-saving machines in the future.

The robot, dubbed “DribbleBot,” is a small, four-legged machine that can deftly move a ball across a variety of terrains, from sand to mud to snow. It can also get up and recover after falling, according to MIT News.

DribbleBot was created at MIT’s Improbable Artificial Intelligence Lab, part of the school’s Computer Science and Artificial Intelligence Laboratory (CSAIL). Pulkit Agrawal, a principal investigator at CSAIL, believes the technology used to create a soccer-playing robot could have wide-ranging implications.

“If you look around today, most robots are wheeled. But imagine that there’s a disaster scenario, flooding, or an earthquake, and we want robots to aid humans in the search-and-rescue process. We need the machines to go over terrains that aren’t flat, and wheeled robots can’t traverse those landscapes,” Pulkit Agrawal said in a statement. “The whole point of studying legged robots is to go [on] terrains outside the reach of current robotic systems.”

Researchers have been actively trying to program robots to play soccer for some time, according to MIT. The team behind DribbleBot wanted their machine to be especially adaptive, learning through trial and error how to navigate real-world terrains. To give the robot these “hard-to-script skills,” researchers turned to a simulation.

A replica of the robot and a ball were loaded into a digital version of the natural world, and researchers were able to set specific physics parameters for the robot to learn from inside the simulation. This method allowed them to simulate 4,000 versions of the robot simultaneously in real time. Therefore, they could collect data 4,000 times faster than if they just used one robot.

DribbleBot actually started without knowing how to dribble a ball at all. But researchers designed a system that gave it positive or negative reinforcement depending on how well it performed the task in the simulation. The machine could then narrow down the correct sequence of motions needed to keep the ball in control.

“In real time, it’s a couple of days, and in the simulator, hundreds of days. Over time it learns to get better and better at manipulating the soccer ball to match the desired velocity,” said MIT Ph.D. student Gabe Margolis in a statement.

Getting a robot to dribble a ball is much more complicated than getting one to simply walk around outside. One of the reasons for this is that the interactions between the ball and the landscape can be different than the interactions between the robot and the landscape, according to MIT. For example, thick grass or snow will slow a ball’s movement, but a robot’s legs will not be as affected by these factors.

“Past approaches simplify the dribbling problem, making a modeling assumption of flat, hard ground. The motion is also designed to be more static; the robot isn’t trying to run and manipulate the ball simultaneously,” Yandong Ji, a research assistant in the Improbable AI Lab, told MIT News.

DribbleBot is equipped with all the high-tech tools it needs in a compact package. It has a set of sensors on its head and body that let it “see” nearby surroundings and an on-board computer “brain” that converts sensor data into actions. The robot is able to feel the nuances of the terrain it is on through leg sensors and adjust its actions accordingly.

There are still ways to improve DribbleBot, according to the team at MIT. For example, it struggles with slopes and stairs, sometimes getting stuck on areas with elevation change. Researchers hope that they will be able to teach the robot how to lift a ball over a step in the future. They will also explore how the technology behind DribbleBot can be applied to other robots, allowing machines to quickly transport a range of objects around outside using legs and arms.