The Improbable Artificial Intelligence Lab at the Massachusetts Institute of Technology (MIT) has developed an innovative robot, dubbed the Dexterous Ball Manipulation with a Legged Robot (DribbleBot), capable of dribbling a soccer ball under real-world conditions. This significant breakthrough in robotics demonstrates the potential for legged robots to operate in complex, uneven environments, making them valuable in search and rescue operations and other challenging scenarios.
Robot soccer, also known as football to some, has been in existence since the mid-1990s. Although these robotic soccer matches have traditionally been simplified versions of the human game, they remain a popular research topic among roboticists. Prior research efforts in this field have generally focused on wheeled robots playing on flat, uniform surfaces, chasing a ball that is allowed to roll to a stop.
In contrast, DribbleBot utilizes a quadruped robot equipped with two fisheye lenses and an onboard computer featuring neural network learning capabilities. This advanced technology enables the robot to track a size 3 soccer ball across a surface mimicking the uneven terrain of a real soccer pitch, complete with sand, mud, and snow. The unpredictable rolling of the ball on this terrain, combined with the risk of falling, challenges the 40-cm (16-in) tall robot to recover and retrieve the ball like a human player.
While the impressive feats of Boston Dynamics robots, such as running on rough terrain and performing backflips, may make DribbleBot’s accomplishments seem simple, the act of dribbling a ball on uneven terrain is significantly more complex. Unlike walking robots that rely on external visual sensors and analyze foot grip to maintain balance, DribbleBot must learn the necessary skills to control a ball on the move, responding to minute factors that do not affect the dribbler.
To expedite the learning process, researchers conducted 4,000 digital simulations of the robot in parallel in real time, incorporating the dynamics and responses to the simulated ball’s movement. The robot received positive reinforcement for successful dribbling and negative reinforcement for errors, allowing hundreds of days of play to be compressed into just a few days. DribbleBot’s onboard camera, sensors, and actuators then allowed it to apply the digital learnings to real-world situations, refining its skills against the more complex reality.
Pulkit Agrawal, MIT professor, CSAIL principal investigator, and director of the Improbable AI Lab, highlights the importance of developing legged robots: “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. The whole point of studying legged robots is to go terrains outside the reach of current robotic systems. Our goal in developing algorithms for legged robots is to provide autonomy in challenging and complex terrains that are currently beyond the reach of robotic systems.”
The groundbreaking research on DribbleBot will be presented at the 2023 IEEE International Conference on Robotics and Automation (ICRA) in London, commencing on May 29, 2023. A video discussing DribbleBot’s capabilities and accomplishments is available for viewing below.