Skip to main content

Scientists Discover Way to Send Information into Black Holes Without Using Energy

3 comments

Humanoid Robots Learn to Play Tennis Like Humans With New Innovative System

For decades, roboticists have faced a significant challenge: teaching humanoid robots to perform athletic sports skills with the same fluidity, precision, and speed as humans. Sports like tennis demand highly dynamic motion, rapid reactions, and accurate control—qualities that traditional robots struggle to achieve. While some progress has been made in areas like table tennis and football, robots have typically been limited in agility and realism, often relying on highly controlled environments or simplified tasks.

Now, researchers in China have developed an innovative system that pushes humanoid robots closer to real-world athletic performance. This new approach, described in a recent preprint paper on arXiv, shows that robots can learn tennis skills effectively even from imperfect, fragmentary human motion data. This represents a shift from previous methods that tried to replicate full human motion, which often proved too complex or physically unfeasible for humanoid robots.

The Challenge of Teaching Sports to Robots

Traditional methods for teaching robots athletic skills often rely on kinematic data or video-based extraction of human movements. While these methods provide detailed insights into human motion, they are typically difficult to implement on robots. Robots have physical limitations that make perfectly replicating human movements impossible, and small errors can lead to unrealistic or unstable behavior. Moreover, collecting high-quality motion data is time-consuming and expensive.

Even when robots have been trained to play sports like table tennis or football, they often display limited agility and unnatural movement. Multi-step actions, rapid reactions, and complex coordination are difficult to achieve without a breakthrough in how robots learn and execute athletic skills.

Introducing the LATENT System

The new solution comes from a collaboration between researchers and the Chinese AI robotics company Galbot. Their project, called the LATENT system, is short for Learns Athletic humanoid TEnnis skills from imperfect human motioN daTa. The key innovation of LATENT is its ability to leverage imperfect human motion data, which makes the learning process simpler and more practical.

Instead of relying on complete, perfect demonstrations of tennis gameplay, LATENT focuses on primitive tennis skills. These include basic movements like forehand and backhand strokes, lateral shuffles, and crossover steps. By working with smaller, fragmented motion data, the system avoids the complexity of full-game human motion, while still providing valuable guidance for the robot.

As the research team explains, “Our key insight is that, despite being imperfect, such quasi-realistic data still provide priors about human primitive skills in tennis scenarios. With further correction and composition, we learn a humanoid policy that can consistently strike incoming balls under a wide range of conditions and return them to target locations, while preserving natural motion styles.”

How LATENT Works

The development of LATENT involved several steps:

  1. Data Collection – Researchers recorded five hours of primitive tennis skill fragments from amateur players using a compact motion capture system. These fragments captured the essential movements necessary for tennis without requiring perfect or complete motions.

  2. Latent Action Space – The team created a “latent action space” from the collected data. This is essentially a representation of the robot’s possible actions, built from human movement fragments. It allows the system to interpolate and adapt motions to different situations.

  3. Reinforcement Learning and Simulation – Using reinforcement learning, the robot practiced in simulations, refining its movements and learning to adapt to different ball trajectories and court positions. This step allowed the system to test and improve its skills safely before real-world deployment.

  4. Real-World Deployment – Finally, the LATENT system was implemented on a Unitree G1 humanoid robot, which began testing its skills on actual tennis courts.

Testing on Real Courts

To evaluate the system, the research team conducted extensive trials where robots played tennis against human players. They focused on forehand and backhand strokes and tested performance across both forecourt and backcourt regions. In total, the researchers analyzed 10,000 trials, comparing LATENT’s performance to previous approaches.

The results were impressive. While the robots could not yet compete with professional tennis players, they were capable of performing multi-shot rallies with humans. The robots could switch between forehand and backhand strokes, adapt to various court regions, and return balls with remarkable precision. In simulations and real-world tests, LATENT outperformed previous methods in success rate, accuracy, and motion naturalness. In its best performance, the robot achieved a 96.5% success rate, defined as successfully returning a ball to within 2.5 meters of the target.

Why LATENT is a Breakthrough

The success of LATENT lies in its approach to using imperfect data. By focusing on fragments of human motion, the system simplifies the learning task while retaining essential priors about human tennis behavior. The latent action space enables the robot to compose these primitive motions into coherent, effective gameplay strategies. Reinforcement learning then fine-tunes these skills, allowing the robot to adapt to different scenarios.

This method offers several advantages over traditional approaches:

  • Efficiency: Collecting fragmentary data is faster and easier than capturing perfect full-game motions.

  • Adaptability: The system can generalize from imperfect data to a range of real-world conditions.

  • Natural Motion: Despite working with imperfect data, the robot’s movements appear fluid and human-like.

Future Directions

The researchers acknowledge that LATENT is not yet perfect. Currently, the system relies on a motion capture system to track the robot’s environment, which may limit deployment in less controlled settings. The team suggests that incorporating active vision could allow the robot to perceive the environment independently, further improving performance.

Additionally, the task formulation focuses on returning randomly initialized balls to target locations, which is simpler than a full two-player tennis match. Future work may involve developing strategies for more complex gameplay, such as predicting an opponent’s movements, responding to different ball spins, and planning multi-shot sequences in real time.

Despite these limitations, LATENT demonstrates a powerful principle: robots can learn complex athletic skills even from imperfect data. This opens doors for applications beyond tennis, potentially extending to other sports or tasks where perfect human motion data is difficult to obtain.

Implications for Robotics and AI

LATENT represents a meaningful step forward in humanoid robotics and artificial intelligence. By showing that imperfect data can still yield highly effective motion policies, the study challenges the notion that robots must exactly replicate human movement to perform human-like tasks. This insight could accelerate the development of robots capable of interacting safely and efficiently with humans in a variety of physical activities, from sports and rehabilitation to household assistance and industrial tasks.

Moreover, the combination of motion fragments, latent action spaces, and reinforcement learning may serve as a template for other robotic learning systems. The approach balances data efficiency, adaptability, and naturalness, three factors that are often difficult to achieve simultaneously in humanoid robotics.

Conclusion

Teaching humanoid robots to play tennis has long been a challenging goal, but the LATENT system demonstrates that even imperfect human motion data can be leveraged to achieve remarkable results. By focusing on primitive skills and building a latent action space, robots can learn to perform dynamic, precise, and natural tennis movements. In trials, LATENT-equipped robots successfully returned balls with high accuracy and participated in multi-shot rallies with human players, surpassing previous methods in both simulation and real-world environments.

While further improvements are needed to approach professional-level performance or full two-player matches, LATENT lays the foundation for a new generation of athletic humanoid robots. Its principles may soon extend beyond tennis, enabling robots to master other sports, complex physical tasks, and interactions that were once considered out of reach.

The research, led by Zhikai Zhang and colleagues, highlights a promising future for robotics where learning from imperfect data can produce practical, efficient, and naturalistic behaviors in humanoid machines. For the first time, robots are not just mimicking human motion—they are learning to play like humans, and the court is just the beginning.

Reference: Zhikai Zhang et al., Learning Athletic Humanoid Tennis Skills from Imperfect Human Motion Data, arXiv (2026). DOI: 10.48550/arxiv.2603.12686


Comments

Post a Comment

Popular

Scientists Discover Way to Send Information into Black Holes Without Using Energy

For years, scientists believed that adding even one qubit (a unit of quantum information) to a black hole needed energy. This was based on the idea that a black hole’s entropy must increase with more information, which means it must gain energy. But a new study by Jonah Kudler-Flam and Geoff Penington changes that thinking. They found that quantum information can be teleported into a black hole without adding energy or increasing entropy . This works through a process called black hole decoherence , where “soft” radiation — very low-energy signals — carry information into the black hole. In their method, the qubit enters the black hole while a new pair of entangled particles (like Hawking radiation) is created. This keeps the total information balanced, so there's no violation of the laws of physics. The energy cost only shows up when information is erased from the outside — these are called zerobits . According to Landauer’s principle, erasing information always needs energy. But ...
3 comments
Read more

Black Holes That Never Dies

Black holes are powerful objects in space with gravity so strong that nothing can escape them. In the 1970s, Stephen Hawking showed that black holes can slowly lose energy by giving off tiny particles. This process is called Hawking radiation . Over time, the black hole gets smaller and hotter, and in the end, it disappears completely. But new research by Menezes and his team shows something different. Using a theory called Loop Quantum Gravity (LQG) , they studied black holes with quantum corrections. In their model, the black hole does not vanish completely. Instead, it stops shrinking when it reaches a very small size. This leftover is called a black hole remnant . They also studied something called grey-body factors , which affect how much energy escapes from a black hole. Their findings show that the black hole cools down and stops losing mass once it reaches a minimum mass . This new model removes the idea of a “singularity” at the center of the black hole and gives us a better ...
Post a Comment
Read more

How Planetary Movements Might Explain Sunspot Cycles and Solar Phenomena

Sunspots, dark patches on the Sun's surface, follow a cycle of increasing and decreasing activity every 11 years. For years, scientists have relied on the dynamo model to explain this cycle. According to this model, the Sun's magnetic field is generated by the movement of plasma and the Sun's rotation. However, this model does not fully explain why the sunspot cycle is sometimes unpredictable. Lauri Jetsu, a researcher, has proposed a new approach. Jetsu’s analysis, using a method called the Discrete Chi-square Method (DCM), suggests that planetary movements, especially those of Earth, Jupiter, and Mercury, play a key role in driving the sunspot cycle. His theory focuses on Flux Transfer Events (FTEs), where the magnetic fields of these planets interact with the Sun’s magnetic field. These interactions could create the sunspots and explain other solar phenomena like the Sun’s magnetic polarity reversing every 11 years. The Sun, our closest star, has been a subject of scient...
Post a Comment
Read more