Tech • IA • Crypto
China's New AI Robot MOYA Feels Too Real (92% Human)
7/10TL;DR
A surge of breakthroughs in humanoid robotics highlights rapid progress across lifelike interaction, industrial automation, autonomous skills, and AI control systems.
KEY POINTS
China’s Moya sparks fascination and unease
Shanghai-based DroidUp unveiled Moya, a humanoid robot designed for lifelike social interaction rather than labor. Standing about 5.5 feet tall and weighing 70 pounds, it features warm silicone skin at 90–97°F, facial tracking cameras, and the ability to mirror human expressions in real time. Its design includes a flexible artificial spine and padded structure mimicking human muscle and fat, aiming to blur the line between machine and person.
Financial and technical skepticism surrounds DroidUp
Despite raising $28.5 million, nearly half has already been spent by the 32-person startup, raising concerns about sustainability. Critics have questioned the realism of Moya’s movements and highlighted inconsistencies in demonstrations, including edited promotional footage. The robot is priced at $173,000, with an initial production run of about 50 units, targeting sectors like elder care, hospitals, and public-facing services.
Boston Dynamics Atlas nears real industrial deployment
Boston Dynamics, owned by Hyundai Motor Group, is pushing humanoid robots toward factory readiness with its Atlas platform. Advanced simulation allows the equivalent of millions of training hours in a single day, with learned behaviors transferable to physical robots in roughly one hour. This drastically shortens development cycles compared to just a few years ago.
Hardware design accelerates Atlas capabilities
Atlas uses only two actuator types and a symmetrical body design, simplifying simulation and control. Eliminating joint cables enables continuous rotation and reduces maintenance. These engineering choices allow Atlas to perform complex tasks such as lifting a 100+ pound refrigerator, exceeding its training range, demonstrating real-world adaptability.
Strategic partnerships drive rapid progress
Collaborations with Google DeepMind and NVIDIA underpin Atlas’s advancements. High-performance computing and AI learning systems are compressing timelines and enabling more sophisticated physical intelligence. Analysts project Boston Dynamics could capture 15% of the global humanoid robot market by 2035, and up to 60% of the premium industrial segment.
Agibot achieves fully autonomous table tennis
Agibot’s Yuan Zhang A3 humanoid robot completed a full table tennis match with no human intervention. Using a 20 kHz spike camera and advanced motion algorithms, the robot processes fast-moving balls exceeding 5 m/s, predicting trajectories with millimeter precision. The system integrates perception, planning, and execution in a continuous loop, marking a major milestone in real-time robotic coordination.
Table tennis emerges as a robotics benchmark
The complexity of table tennis—requiring rapid reaction, spin prediction, and dynamic movement—makes it a demanding test for humanoid systems. Agibot’s success demonstrates progress in closed-loop autonomy, where robots independently adapt to unpredictable environments without pauses or external control.
Cody targets child-friendly human-robot interaction
Seattle-based Mind Children introduced Cody, a 3-foot-tall humanoid designed for museums, hotels, and educational spaces. With expressive animations and autonomous navigation, Cody focuses on engagement and accessibility rather than physical strength. The robot operates without human control, using onboard AI for interaction and decision-making.
Decentralized AI powers Cody’s autonomy
Cody runs on Hyperon, a decentralized AGI framework from SingularityNET, emphasizing reasoning, adaptive behavior, and goal prioritization. The project has raised over $600,000 through crowdfunding toward a $1 million target, with future plans including healthcare and education deployments and expanded physical capabilities.
Alibaba launches Qwen Robot AI ecosystem
Alibaba introduced Qwen Robot, a suite of embodied AI models designed to bridge the gap between perception and physical action. The system is divided into three components: Qwen-Nav for navigation, Qwen-Manip for object handling, and Qwen-World for predictive reasoning about environments.
Real-world performance and open ecosystem strategy
In testing, Qwen-Nav guided a robot through an unfamiliar apartment using only a single low-resolution camera with 196 ms latency. Qwen-Manip, trained on 38,000 hours of data, achieved a 59.83 process score and 45% task success rate in benchmarking. Alibaba also released Chat2Robot, an open platform to encourage development and integration.
Global competition intensifies in robotics ecosystems
The United States leads in AI research through firms like DeepMind and NVIDIA, while China dominates hardware manufacturing with companies such as Unitree and Agibot. Alibaba’s strategy positions Qwen Robot as a unifying AI layer to connect China’s hardware ecosystem into a cohesive platform.
CONCLUSION
Recent developments show humanoid robotics advancing simultaneously in realism, autonomy, and scalable AI infrastructure, signaling a transition from experimental prototypes to practical deployment across industries.