PingpongDuo Lets Humanoid Robots Play Table Tennis Doubles
Researchers have introduced PingpongDuo, a robotic framework enabling humanoid robots to play table tennis doubles through organised movements and alternating strikes. The system addresses challenges inherent in co-operative robotic play, including sharing limited space, swift repositioning, striking accuracy, and avoiding physical contact between partners.
The framework integrates coordinated positioning with learned movement and racket control, allowing the robots to adapt to varied ball trajectories while operating as a team. Experiments demonstrated improved ball return performance in simulated environments. It confirmed that two humanoid robots could maintain consecutive rallies, adhering to the alternating-strike rule required in doubles play.
The University of Hong Kong, Shanghai Innovation Institute, Archon Robotics, and Big Tech company NVIDIA collaborated on the PingpongDuo development. This control framework facilitates two humanoid robots playing table tennis doubles, coordinating their movements, and alternating shots. And this differs markedly from previous systems.
The system combines artificial intelligence, motion learning, and risk estimation, assisting robots in striking the ball, vacating space for teammates, and repositioning for subsequent returns without collision. Unlike conventional humanoid table tennis systems designed for individual play, PingpongDuo coordinates two robots within the confined space on one side of the table.
Each robot must strike in strict alternation, then move away from the central hitting area, allowing its teammate to prepare for the next return. The framework operates on a hierarchical control system, combining team-level positioning with individual movement and striking skills. So, at the team level, a planner assigns positioning targets for both robots.
A learned risk estimator assesses potential movements for collision risks involving their hands, feet, and bodies. A safety filter then adjusts any potentially risky positioning plans, prioritising the robot preparing to strike to preserve its ability to return the ball. At the individual level, the system employs distinct artificial intelligence policies for footwork and racket control.
The researchers trained the movement policy by using human table tennis doubles footage, compiling a library of approximately 200 footwork sequences covering various movement distances. The robots acquire the capacity to transition between striking, clearing space, recovering, and waiting. But training each skill using the body positions and movements generated by the other robot assists them in responding to new shots without reverting to a standard posture first.
According to Interesting Engineering, the researchers also devised a racket arm refinement policy to enhance ball returns following rapid movements. This policy adjusts seven arm joints based on the robot's current state and the estimated position and velocity of the incoming ball.
To train this specific component, the team calibrated a physics simulator using actual measurements of ball flight, table bounces, and racket contact. The resulting model permits the system to ascertain how variations in racket movement affect whether a ball successfully clears the net and lands on the opponent's side.
In simulation, PingpongDuo enhanced ball return success by 13.2 per cent when compared to the established baseline. The researchers also reported a positioning error of approximately 2 centimetres during the clearance phase. And these figures represent a notable step in robotic co-ordination.
Real-world experiments involved two Unitree G1 humanoid robots, which sustained more than 10 consecutive alternating strikes. One robot notably completed a 2.15-meter clearance and recovery movement cycle in 2.4 seconds, achieving a peak speed of 2.23 metres per second.
The researchers noted that the current iteration of the system relies on external motion capture equipment and does not account for ball spin. Future work intends to explore onboard perception, spin estimation, and the development of autonomous attacking and defensive strategies.
A new robotic framework, PingpongDuo, enables humanoid robots to play table tennis doubles.
The system allows two robots to coordinate movements, alternate strikes, and avoid collisions.
Experiments showed improved ball-return rates and sustained rallies of over 10 consecutive strikes.
Future work will focus on onboard perception, ball spin, and autonomous strategy development.
Source: Interesting Engineering
