MIT’s Improbable Artificial Intelligence Lab has developed dexterous ball manipulation using a legged robot (DribbleBot) that can dribble a soccer ball under real-world conditions similar to those encountered by human players.
Robot soccer (football for some) has been around since the mid-1990s, but these games tend to be much simplified versions of the human game. But having a robot control a ball is also a very attractive research topic for roboticists.
Typically, these research efforts focus on wheeled robots chasing a ball that rolls to rest on a very flat, uniform surface. In the case of DribbleBot, the team uses a quadrupedal robot with two fisheye lenses and a neural network learning function to track a size 3 soccer ball over uneven areas of a real pitch, including sand, mud and snow. used his computer onboard. Not only did this make it harder to predict when the ball would roll, it also increased the risk of falling. A 40-cm (16-inch) tall robot had to retrieve and retrieve the ball like a human player.
and
This might seem easy in a world of Boston Dynamics robots scurrying across broken ground and doing backflips, but dribbling makes all the difference. Walking robots can rely on external visual sensors and rely on analyzing how firmly their feet are gripping the ground to maintain balance. A ball rolling over uneven terrain is much more complicated as it reacts to small factors that have no effect on the dribbler. The robot must discover on its own the skills necessary to control the ball while both are in motion.
To speed up this process, 4,000 digital simulations of the robot were run in parallel in real-time, including the dynamics involved and how it would react to the simulated roll of the ball. Positive reinforcement was given when the robot learned to dribble the ball, and negative reinforcement if it made an error. These simulations allowed us to compress hundreds of days of play into just a few days.
Then, in the real world, the cameras, sensors, and actuators on board the robots allowed them to apply what they learned digitally and hone those skills to more complex realities.
and
Pulkit Agrawal, professor at MIT, principal investigator at CSAIL, and director of the Improbable AI Lab, said: “But imagine there is a disaster scenario, a flood, or an earthquake, and robots need to assist humans in the search and rescue process. I can’t.” The whole point of researching legged robots is to go to terrains outside the range of current robotic systems. It’s about providing autonomy in challenging and complex terrain that is beyond…”
The research will be presented at the 2023 IEEE International Conference on Robotics and Automation (ICRA) in London on May 29, 2023.
The video below explains the DribbleBot.
dribble bot
Source: MIT
