AI Robot Table Tennis Tech News Is Wild (UNITREE G1 CODROID 2)

➡ Unitree’s G1, a table tennis-playing robot, uses advanced AI to predict ball trajectories and make precise movements, even outperforming humans in some cases. Aston Codroid’s Codroid II robot, designed for factory settings, has high load capacity and response speed, and can interact with humans and other machines. VMR’s Omnihead, a modular head for robots, provides 360° vision and sound localization, enabling autonomous navigation and interaction. Lastly, Vin Motion’s Motion 1 robot can perform light industrial tasks, and Tencent’s Voyager can create 3D scenes from a single photo, while a new AI model, 1.2.2S2V, enhances audio-driven human animation.

What if your general-purpose AI robot assistant could outsmart, outcreate, outplay and even outrun you? Well now, with Unitree’s G1, it’s finally happening thanks to a breakthrough AI system enabling rapid coordinated reactions to fast-moving objects. With ball speeds often exceeding 5 meters per second, table tennis demands sub-second perception, prediction and action, an area where robots have struggled to match human agility and precision until now. Enter humanoid table tennis robot, a unique hierarchical framework to combine model-based planning and reinforcement learning driven whole body control for humanoid table tennis. In fact, its secret lies in a model-based planner that precisely forecasts the ball’s trajectory, then selects the ideal racket position, movement speed and timing for each strike.

Next, a reinforcement learning-based controller translates these plans into smooth, human-like movements, coordinating both the robot’s arms and legs to deliver natural strikes while maintaining overall stability, even during rapid, back-to-back rallies. And to ensure the robot not only plays effectively, but also moves convincingly, the system incorporates reference data from actual human motion during its training process. This allows the robot to replicate the subtle nuances of human strikes and postures for more lifelike and engaging matches, and to validate the system in the real world with a general-purpose humanoid, the framework was put to the test against both human and robot opponents.

The results were that Unitree’s G1 sustained up to 106 consecutive shots in rallies when playing a human, and continuous shots when playing with another robot. But what task or skill should this tech be applied to next in order to achieve split-second reactions while maintaining coordination? And by how much should robots be able to outperform humans before their superiority becomes unsettling, unsafe, or unwise for us? Already, robots are preparing to outwork us, as Aston Codroid just unveiled its latest batch of new humanoids. What’s known so far about the company’s Codroid II robot is that it’s 170 cm tall and weighs 70 kg without its battery installed.

As for strength, each of the robot’s arms can lift a payload weighing up to 5 kg, with a total arm length of 602 mm. On top of this, Codroid II’s body features between 29 to 31 degrees of freedom, depending on the robot’s configuration, enabling a broad range of movement for handling complex tasks. But what really sets the Codroid II apart is its focus on in-house innovation, with its integrated joint modules, including a stiff actuator and a proprietary proprioceptive actuator, having been developed entirely by Aston Codroid’s team. Even more importantly though, the robot’s stiff actuator delivers higher load capacity and a wider gear ratio range, all while maintaining a compact, lightweight design with minimal noise.

On top of this, the proprioceptive actuator provides high response speed, high torque density, and notable impact resistance, which are crucial for factory settings. Furthermore, the Codroid II is equipped with an embodied intelligence agent that leverages high and low frequency system synergy, a design aimed at improving real-time interaction and coordination with both human operators and other machines. In fact, Codroid II’s rollout coincides with rising momentum around AI in China, and the delivery of these humanoid robots is seen as a direct answer to the government’s latest action plan supporting artificial intelligence applications. With this batch, Aston Codroid is bridging the gap between current automation and smart factories of tomorrow.

As for the robot’s commercial release, Codroid II units will soon be at work in customers’ factories across various sectors, collaborating with existing industrial and mobile robots to enhance automation in handling, sorting, and assembly processes. But what if robots had modular heads that you could pop on and off to customise your Android’s hardware setup? Finally, VMR just officially introduced its Omnihead, the world’s first dedicated head module for humanoids. Until now, robots have struggled with limited environmental awareness and often depended on remote operators, but Omnihead looks to change this dynamic entirely. At its core, Omnihead features a trio of RGBD camera arrays, delivering a panoramic 360° by 90° field of view.

This enables robots not only to see their surroundings in full, but also to localise themselves, autonomously navigate, and reconstruct three-dimensional environments, even in complex scenarios. Plus, with deep integration of large AI models and a 6-microphone circular array, the head module supports advanced features like 360° sound localisation, multi-turn dialogue, and emotional interaction in both English and Chinese. Omnihead’s open interfaces even allow for the fusion of visual and audio data, making it compatible with multiple humanoid platforms. As a result, it could potentially serve as a general-purpose piece of hardware for enthusiasts as well as businesses. By providing robots with human-like perception and cognition, Omnihead is paving the way for autonomous robots to engage with real-world environments, no teleoperation required.

But there’s another player coming to the AI hardware market. As Vietnam’s Vin Motion just introduced its new Motion 1 robot, its first humanoid robot prototype designed for light industrial tasks including warehouse transport, visual inspection, and basic assembly support. Motion 1 can already walk, wave, recognise objects in videos, and respond with basic gestures or even voice commands. On top of this, Vin Motion is developing SuperMotion, its next-generation humanoid robot aimed at handling heavier industrial applications in the near future. And in the latest generative AI breakthroughs, Tencent just announced Voyager, its new AI system to generate spatially consistent three-dimensional scenes from a single photograph, without relying on traditional modelling workflows.

Users simply upload an image and outline a virtual camera path. Voyager then produces a continuous video that simulates camera movement through the newly generated scene. This approach aims to make virtual three-dimensional environment creation accessible, eliminating the need for manual modelling or technical expertise. But the heart of Voyager actually uses the RGBD technique by combining colour and depth information to accurately estimate distances and maintain object consistency even when viewing from unusual angles. Plus, it features a memory-efficient world cache that stores scene regions and recalls hidden areas as needed, optimizing memory and supporting longer, smoother camera paths. On top of this, Voyager is trained on a massive dataset of real-world videos and virtual scenes, learning how cameras move and how objects appear from all directions.

The system can also directly output three-dimensional reconstructions, such as point clouds or Gaussian proxies. Tencent says at least 60 GB of GPU memory is required for 540-pixel output, and the code is now publicly available for developers. And finally, a new open-source AI model called 1.2.2S2V was just unveiled for the next generation of audio-driven human animation. Featuring a 14 billion-parameter model, 1.2.2S2V is designed to move beyond the limitations of basic talking head technology. Because of this, the approach means that long video dynamic consistency can be maintained throughout entire scenes, which is crucial for professional storytelling and immersive experiences.

On top of this, 1.2.2S2V offers advanced motion and environment control through simple instructions, giving creators precise authority over not just character movement but also the surrounding environment. Anyways, like and subscribe for more of the latest AI news, and check out this video here. [tr:trw].