A Robotic Table Tennis System Tracks The Ball’s Trajectory And Modifies Its Swing In Real Time

Over the past decades, roboticists have developed various systems capable of accurately mimicking specific human movements and behaviors. Some robots can compete against humans or others in sports, like at RoboCup, a robot soccer competition.

MIT Develops Precise Robotic Table Tennis System

MIT researchers developed a robotic table tennis system that quickly and accurately hits balls. The platform replicates diverse hitting styles and precise spins, as detailed in an arXiv paper.

“The Biomimetic Robotics Lab at MIT builds high-performing robots by advancing hardware and control,” said Kendrick Cancio.

“We developed this table tennis system to study dynamic manipulation and aim for human-level play on a humanoid robot.”

The Biomimetic Robotics Lab researches dynamic legged locomotion and rapid object manipulation. Each area presents distinct challenges.

The Control Challenge of Robotic Table Tennis

Legged locomotion struggles with environmental disturbances, while object manipulation requires precise maneuvers.

“Table tennis demands adaptability and precision as new ball data arrives,” said David Nguyen. “This creates a unique control problem that we believe our custom robotics hardware is well-equipped to handle exceptionally well.”

The new platform created by Nguyen, Cancio, and Sangbae Kim features a robotic arm paired with a control algorithm. The algorithm predicts the ball’s path and plans the arm’s swing to meet strike conditions.

“During the swing, the arm updates its path to hit the ball precisely,” Nguyen explained.

“We found planning the full swing is more reliable but requires more aggressive arm movements.” This lets us push our custom hardware beyond standard system limits.

Their robotic system has two main parts: the perception and actuation modules. The perception module uses motion tracking to locate the team’s custom table tennis balls.

“We predict the ball’s trajectory to determine the expected strike location and timing,” Cancio added.

“At the same time, we solve a nonlinear optimization problem that uses these values, along with our desired strike parameters, to generate a swing trajectory for the arm. Our model predictive controller continuously updates this trajectory, allowing the arm to respond as it receives new ball position data.”

Custom MIT Humanoid Arm Enables Fast, Adaptive Table Tennis Play

The robotic arm used in the team’s table tennis system is a customized version of a humanoid arm developed at MIT. It features high torque and low rotor inertia, enabling it to swing rapidly while quickly adjusting its movements if initial ball trajectory predictions are off.

“We demonstrate the ability to adapt a trajectory to accurately intercept a moving object,” said Nguyen. “While table tennis may not save lives, this type of control could be vital in challenging search and rescue missions where a versatile robot, like a humanoid, must intercept objects.”

Nguyen, Cancio, and Kim tested their robotic table tennis platform in real-world experiments and found it performed exceptionally well. In these initial trials, the robotic arm successfully returned incoming balls 88% of the time, achieving an average exit velocity of 11 m/s and executing three different hitting styles.

Bridging Model-Based and Reinforcement Learning in Robotics

“Currently, robotics tends to be divided between model-based methods and reinforcement learning, with many expecting reinforcement learning to become the dominant approach soon,” said Cancio. “Our work shows that constraint-based optimization still plays a vital role in high-performance systems, and we aim to combine the strengths of both methods when appropriate.”

The system developed by this research team could soon inspire other roboticists to create similar automated table tennis platforms. Additionally, Nguyen, Cancio, and Kim hope to apply their hardware and control algorithms to other dynamic manipulation tasks.

“Since submitting the paper in September, we’ve made significant improvements,” Nguyen added. “We can now aim at specific spots on the table and plan the full trajectory and contact between paddle and ball.”

Looking ahead, the researchers plan to further enhance the robotic table tennis platform. For example, by extending the MIT humanoid arm’s workspace with a gantry—a supporting structure—they could enable it to play complete matches against human opponents.

“We want to keep improving our system’s performance by expanding the workspace with a gantry and greatly increasing ball exit speeds,” Cancio said.

“We also aim to track standard table tennis balls in the future to better compare our robot’s performance against humans and other robotic systems.”

Read the original article on: Tech Xplore

Read more: Goodbye Human Coaches The Future of Tennis May Be in This Robot’s Hands