Table Ping Pong Robot Delivers Swift, Precise Returns

MIT engineers have entered the robotic ping pong scene with a lightweight, high-performance design capable of returning shots with rapid precision.

The new table tennis robot features a multi-jointed arm mounted at one end of the ping pong table, equipped with a regular paddle. Using multiple high-speed cameras and an advanced predictive control system, it rapidly calculates the incoming ball’s speed and path, then performs one of several stroke types—such as loop, drive, or chop—to accurately return the ball to a target spot on the table with controlled spin.

During testing, engineers launched 150 balls at the robot in succession from the opposite side of the ping pong table. The robot achieved an overall return rate of approximately 88% across all three swing styles. Its striking speed rivals the fastest returns made by human players and surpasses that of other robotic table tennis systems.

Expanding Robot’s Range of Motion for Greater Shot Variety and Market Potential

The team now aims to expand the robot’s range of motion so it can handle a broader variety of shots. With this improvement, they see potential for the system to become a strong contender in the growing market of intelligent robotic training tools.

Beyond table tennis, the researchers believe the technology could enhance the speed and agility of humanoid robots, particularly in high-stakes scenarios like search-and-rescue operations, where rapid and precise reactions are crucial.

“The challenges we’re tackling—especially those involving fast, accurate object interception—could be valuable in situations where a robot needs to perform quick, dynamic movements and determine where its end effector should meet an object in real time,” explains MIT graduate student David Nguyen.

Nguyen co-authored the study with fellow MIT graduate student Kendrick Cancio and Sangbae Kim, an associate professor of mechanical engineering and director of the MIT Biomimetics Robotics Lab. They will present their findings this month at the IEEE International Conference on Robotics and Automation (ICRA).

The Challenge of Building Ping Pong Robots Since the 1980s

Since the 1980s, researchers have been tackling the complex challenge of building robots capable of playing ping pong—a task that demands a unique mix of technologies, including high-speed vision, responsive motors and actuators, precise control of robotic arms, accurate real-time prediction, and strategic game planning.

“As far as control problems go in robotics, you can think of it as a spectrum,” explains Nguyen. “On one end, there’s manipulation—typically slow and highly precise, like carefully grasping an object. On the other, there’s locomotion, which is about dynamic movement and reacting to disturbances. Ping pong falls right in the middle: it requires the precision of manipulation, but under the speed constraints of dynamic tasks—like returning a ball in just 300 milliseconds.”

Robotic ping pong systems have significantly evolved, with recent breakthroughs from companies like Omron and Google DeepMind using AI to learn from past gameplay and adapt to a broader range of strokes and shot types. These advanced systems are now fast and accurate enough to rally with mid-level human players.

For their latest design, the researchers modified a high-power, lightweight robotic arm originally developed by Kim’s lab as part of the MIT Humanoid project—a bipedal robot roughly the size of a small child. This humanoid robot is being used to test a range of dynamic maneuvers, such as navigating uneven terrain, jumping, running, and even performing backflips, with the goal of deploying similar robots for search-and-rescue missions in the future.

Enhancing Robotic Arms for Ping Pong: Adding Precision with Extra Degrees of Freedom

Each arm of the humanoid features four joints, or degrees of freedom, controlled by electrical motors. Cancio, Nguyen, and Kim created a similar robotic arm and adapted it for ping pong by adding an extra degree of freedom in the wrist, allowing for more precise paddle control.

The team mounted the robotic arm at one end of a standard ping pong table and set up high-speed motion capture cameras around it to track balls bounced toward the robot. They also developed advanced control algorithms that use principles of mathematics and physics to predict the necessary speed and paddle angle required to return the ball with a specific type of swing—such as loop (topspin), drive (straight-on), or chop (backspin).

These algorithms were implemented across three computers that simultaneously processed camera data, estimated the ball’s real-time position, and converted these predictions into commands that enabled the robot’s motors to react swiftly and execute the appropriate swing.

After bouncing 150 balls consecutively at the robot’s arm, the researchers found that its accuracy in returning the ball was similar across all three swing types: 88.4% for loop strikes, 89.2% for chops, and 87.5% for drives. They then fine-tuned the robot’s reaction time and discovered that its arm hits balls faster than current systems, reaching speeds of 20 meters per second.

Advancements Toward Human-Level Performance in Ping Pong Robots

In their published study, the team reports that the average strike speed, or the speed at which the paddle makes contact with the ball, is 11 meters per second. Advanced human players typically return balls at speeds between 21 and 25 meters per second. Since their initial experiments, the researchers have continued to refine the system, achieving strike speeds of up to 19 meters per second (about 42 miles per hour).

“One of the main goals of this project is to show that we can reach the same level of athleticism as humans,” says Nguyen. “And in terms of strike speed, we’re getting really close.”

Their follow-up research has also enabled the robot to aim. The team integrated control algorithms into the system, allowing the robot to predict not only how to hit the ball but also where to place it. With this latest update, they can set a target location on the table, and the robot will return the ball to that exact spot.

Because the robot is fixed to the table, it has limited mobility and can mostly return balls that fall within a crescent-shaped area near the table’s midline. In the future, the team plans to mount the robot on a gantry or wheeled platform, which will allow it to cover more of the table and return a wider range of shots.

“One of the key aspects of table tennis is predicting the spin and trajectory of the ball, based on how your opponent hits it—information that an automatic ball launcher can’t provide,” says Cancio. “A robot like this could replicate the movements of an opponent in a game, helping humans to practice and improve.”

Read the original article on: Techxplore

Read more: Why Did This Robot Go Haywire And Endanger Its Handlers?