MIT engineers are in the robotic Ping Pong game with a powerful, delicate design that returns shots with great precision.
The up-to-date table tennis bot contains a multilateral robotic arm, which is set at one end of the ping pong table and has a standard ping pong pong. Supported by several brisk cameras and a high capacity predictive control system, the robot quickly estimates the speed and trajectory of the incoming ball and performs one of several types of swings-pass-pass, drive or koter-a good to hit the ball thoroughly at the desired location on the table with different types of spin.
In tests, engineers threw 150 balls at the robot, one by one, through the ping pong table. Bot successfully returned the balls at a speed of a hit of about 88 percent in all three types of swing. The speed of the robot is approaching the highest speed of the return of human players and is faster than in other robotic tennis projects.
Now the team wants to raise the ray of the robot game so that they can return a wider variety of shots. Then they imagine that configuration can be a profitable competitor in the developing field of knowledgeable robotic training systems.
In addition to the game, the team claims that table tennis technology can be adapted to improve the speed and reaction of humanoid robots, especially in the case of search and rescue scenarios, as well as situations in which the robot would have to react or predict quickly.
“The problems that we solve, especially related to the capture of objects really quickly and accurately, can potentially be useful in scenarios in which the robot must carry out dynamic maneuvers and plan in which his final effector will meet the subject, in real time,” says a graduate student of Mit David Nguyen.
NGUYEN is a co -author of the up-to-date study, together with the graduate of Mit Kendrick Cancio and Sangbae Kim, professor of mechanical engineering and boss Myth of biomimetics robotics lab. Scientists will present the results of these experiments in paper At the IEEE international conference on robotics and automation (ICRA) this month.
Precise game
Building robots for playing ping pong is a challenge that scientists have taken up since the 1980s. The problem requires a unique combination of technology, including brisk machine vision, brisk and agile engines and actuators, precise manipulator control and precise prediction in real time, as well as planning the game strategy.
“If you are thinking about a spectrum of control problems in robotics, we have one final manipulation, which is usually slow and very precise, for example, lifting the object and making sure you grab it well. On the other hand, you have a locomotion, which involves dynamics and adapting to disorders in your system,” explains Nguyen. “Ping Pong sits between them. You are still manipulated because you must be precise in hitting the ball, but you must hit it within 300 milliseconds. It will balance similar problems of dynamic transport and precise manipulation.”
Ping Pong works have gone through a long way since the 1980s, recently with Omron and Google Deepmind projects that operate artificial intelligence techniques to “learn” from previous Ping Pong data to improve the work performance in relation to the growing diversity of strokes and shots. It has been shown that these projects are brisk and precise to gather with indirect human players.
“These are really specialized robots designed to play ping pong,” says Cancio. “Thanks to our robot, we examine how the techniques used in playing ping pong can translate into a more generalized system, such as humanoid or anthropomorphic robot, which can do many different, useful things.”
Game control
Due to their up-to-date project, scientists modified a delicate, robotic arm of high power, which Kim laboratory developed as part of the myth of humanoid-use, duplicate robot, which is more or less the size of a petite child. The group uses the robot to test various animated maneuvers, including navigation in uneven and different terrain, as well as jumping, running and backflyal, to one day of implementing such robots for search and rescue operations.
Each humanoid arms have four joints or degrees of freedom, each of which is controlled by an electric motor. Cancio, Nguyen and Kim built a similar robotic arm that they adapted to Ping Pong, adding an additional degree of freedom in the wrist to allow the paddle to be controlled.
The band set the robotic arm on the table at one end of the standard Ping Pong table and put on quick cameras to capture traffic around the table to follow the balls that are reflected in the robot. They also developed optimal control algorithms that they provide, based on the principles of mathematics and physics, what speed and orientation for rowing should be made to hit the coming ball with a specific type of swing: loop (or topspin), drive (straight) or chop (backspin).
They implemented algorithms using three computers, which simultaneously processed the images of the camera, estimated the state in real time, and translated these estimates to the instructions of the robot engines to react quickly and make swings.
After another bounce of 150 balls in the shoulder, they found a robot hit indicator or the accuracy of the ball’s return, it was more or less the same for all three types of swings: 88.4 percent for loop strikes, 89.2 percent for cutlets and 87.5 percent for disks. Since then, they tuned the reaction time of the robot and found the arm hit the balls faster than the existing systems, at 20 meters per second.
In their article, the team informs that the speed of the robot’s impact or the speed at which the paddle hits the ball is on average 11 meters per second. Advanced human players return balls at a speed of 21 to 25 meters a second. Since the results of their initial experiments, scientists have further improved the system and registered the speed of the impact up to 19 meters per second (about 42 miles per hour).
“Some goals of this project is to say that we can reach the same level of athleticism that people have,” says Nguyen. “In terms of impact speed, we are approaching really, very close.”
Their further work also made it possible to aim the robot. The team has included control algorithms into the system, which not only provides for the way to hit the incoming ball. Thanks to their latest iteration, scientists can set the target location on the table, and the robot will hit the ball to the same place.
Because it is attached to the table, the robot has restricted mobility and range, and can mainly return the balls that reach the crescent -shaped area around the middle of the table. In the future, engineers plan to falsify the bot on porch or circular platform, enabling it to cover more tables and return a wider range of shots.
“The great thing in table tennis is to anticipate the spin and trajectory of the ball, taking into account how the opponent hit her, which is information that the automatic ball launcher will not give you,” says Cantcio. “Such a robot can imitate maneuvers that the opponent would do in the game environment in a way that helps people play and improve.”
These studies are partly supported by Robotics and the AI Institute.