China’s new AI framework helps humanoid robots learn tennis using imperfect human motion, boosting speed, precision, and agility.
Researchers in China have developed a new system that significantly improves how humanoid robots learn to play tennis, marking a step forward in robotic athletic performance. The project, created with Chinese AI robotics firm Galbot , uses a method called LATENT to train robots using imperfect human motion data.
The system works by breaking movements into simple elements like strokes and footwork, making learning more efficient. According to the team, it addresses long-standing challenges in replicating fast, precise, and dynamic sports skills that earlier robotic training methods struggled to achieve.“Our method achieves surprising results in the real world and can stably sustain multi-shot rallies with human players,” said the team in the research abstract. In January 2026, UBTech Robotics’s Walker S2 showed real-world tennis skills, combining perception, balance, and precision, delivering powerful, accurate strokes in human-robot rally demonstrations.Tennis skills unlocked for humanoid robotResearchers in China have developed a new approach to help humanoid robots learn complex tennis skills using imperfect human motion data. Working with Galbot, the team created the LATENT system, which breaks tennis into basic movement fragments such as forehand and backhand strokes, lateral shuffles, and crossover steps.Instead of relying on perfect motion capture or detailed kinematic data, the system uses “quasi-realistic” inputs from amateur players. The researchers collected approximately five hours of these primitive motion fragments using a compact motion-capture setup. Although imperfect, this data provides essential insights into fundamental human tennis movements, reports TechXplore.The team then constructed a latent action space, enabling the robot to interpret, refine, and combine these movements effectively. Using reinforcement learning and large-scale simulations, the system learned how to respond to incoming balls under varied conditions while maintaining natural motion patterns.The trained model was successfully deployed on the Unitree G1 humanoid robot, demonstrating consistent ball striking and targeted returns. The approach addresses long-standing challenges in robotic sports training, particularly the difficulty of replicating fast, dynamic, and precise human athletic behavior with limited or imperfect data.Humanlike robot playResearchers tested the LATENT system in real-world matches, where humanoid robots played tennis against humans across both forecourt and backcourt areas. Evaluations covering 10,000 trials showed strong performance in forehand and backhand strokes, with the system outperforming earlier methods in accuracy, success rate, and motion naturalness. At its peak, the robot achieved a 96.5 percent success rate, consistently returning balls close to target locations.Although not yet capable of matching professional players, the robot demonstrated the ability to sustain multi-shot rallies and adapt to different play conditions. The researchers highlight several areas for improvement in the current system. At present, it depends on motion capture technology for real-world operation, which could be replaced or enhanced with active vision to improve autonomy. The existing setup also simplifies gameplay by focusing on returning randomly generated incoming balls to target locations, rather than simulating a true competitive match. To reach performance levels closer to professional human players, the team suggests adopting a multi-agent training framework. This would allow robots to engage in more realistic, interactive gameplay scenarios and better handle the complexity, strategy, and responsiveness required in actual tennis matches.“Although this work primarily focuses on the tennis return task, the proposed framework has the potential to generalize to a broader range of tasks where complete and high-quality human motion data are unavailable ,” said the team in their research paper.
AI Framework China Galbot Humanoid Robotics Tennis Robot Ubtech Robotics Unitree Unitree G1
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Woman hospitalized after being startled by humanoid robot: 'Are you crazy?'Video widely shared online showed a woman angrily confronting a robot as it waved its metallic arms at her, while a crowd of onlookers gathered around.
Read more »
Chinese humanoid robots could soon beat the fastest human ever in sprinting: ReportChinese humanoid robots could soon surpass human sprint speeds, with experts predicting 100m runs despite key technical hurdles.
Read more »
Watch McDonald’s test humanoid robots on the front lineTech Product Reviews, How To, Best Ofs, deals and Advice
Read more »
UBTech’s goal to mass-produce 10,000 humanoid robots by 2026 gets Siemens backingUBTech partners with Siemens to accelerate humanoid robot production, targeting 10,000 units annually by 2026 amid rising global demand.
Read more »
Chinese lab claims first humanoid robot control using space-based satellite inferenceA Chinese laboratory has reportedly demonstrated the control of a humanoid robot via space-based computing.
Read more »
Muhammad Harfoush on the Animal–Human Bond and Its Expanding Role in Human Well-BeingThere is no shortage of tools marketed to improve human well-being. Medications, nutritional frameworks, psychotherapy models, performance systems, and behavioral theories all reflect a sustained effort to help individuals become healthier and more emotionally resilient.
Read more »