Gasgoo Munich- The 2026 World Robot Conference (WRC) kicked off in Beijing on August 19, with Anyverse Dynamics taking the wraps off a suite of proprietary innovations. The lineup featured the MWA™ Embodied General Brain, the K15 robot, the AnySense Ego first-person multimodal terminal, the 7-DOF Anyverse ADA humanoid arm, and the Anyverse ASC central computing platform.
Beyond the hardware showcase, the company teamed up with HOLLYS—the coffee chain under Korea's KG Group—to build an "Immersive Human-Robot Symbiosis Coffee Space." Through two specific challenges, "Three-Machine Intelligent Collaboration" and "Fingertip Bead Threading," Anyverse Dynamics demonstrated the general brain's capacity for generalization and execution in complex tasks like long-horizon multi-robot coordination and delicate manipulation.
MWA™ Drives Embodied Intelligence into Open Commercial Spaces
Inside the customer-facing zone of the coffee space, Anyverse Dynamics deployed its K15 robot, powered by the MWA™ latent space world model. The robot seamlessly executed a complex service loop: delivering drinks, clearing tables, collecting waste, sanitizing hands, and resetting seats.
Anyverse Dynamics' MWA™ latent space world model does more than just process data; it enables robots to grasp the causal relationship between their actions and changes in the physical world. This builds a core capability for environmental perception and future-state prediction. By leveraging reinforcement learning for high-frequency trial and error during real-world interactions, the system translates a deep understanding of physical laws into precise, safe operational strategies.
Building on that, the company introduced a "long-term bidirectional physical causal chain." Using a time-series Chunk-level inverse dynamics modeling mechanism, the system infers and outputs continuous multi-step Latent Action Chunks. This approach allows for the deduction of long-horizon continuous action chains, fundamentally solving the traditional world model's struggle with error accumulation and action fragmentation during prolonged operations.
The project served as a validation ground for four core capabilities:
Multi-role human-robot symbiosis: Robots are no longer isolated executors. Instead, they accurately interpret human intent, respond to immediate needs, and safely integrate into real-world workflows to collaborate efficiently with people.
Business-oriented long-horizon task planning: The system deeply analyzes service intent, breaking down macro goals into continuous sub-tasks. When faced with changes or environmental interference, it possesses the ability to perform task-level error correction and dynamic re-sequencing.
Dexterous manipulation in complex physical interactions: Handling irregular items like cups, liquids, and flexible cleaning tools, the robot achieves stable grasping, precise placement, and force-controlled operation. This ensures both efficiency and absolute safety in close-proximity human-robot environments.
Dynamic perception in open environments: The system continuously monitors the shifting dynamics of people, objects, and spatial relationships. In unstructured settings, it precisely identifies operational targets and potential risks, replanning behavior in real-time based on on-site conditions.
At this year's WRC, Anyverse Dynamics also showcased the "Three-Machine Intelligent Collaboration" and "Fingertip Bead Threading" tasks, further proving the general brain's robust adaptability across diverse scenarios, from industrial to domestic settings.
In the "Three-Machine Intelligent Collaboration" task, three robots, running on a single model, autonomously divided labor and coordinated across folding, sorting, and packaging stages. The system orchestrated objectives, linked processes, and tracked status in real time—demonstrating comprehensive skills in long-horizon planning, multi-robot collaboration, and anomaly recovery when delays or deviations occurred.
"Fingertip Bead Threading" focused on fine manipulation with a tiny contact space and near-zero margin for error. The robots had to accurately identify and continuously execute millimeter-level grasping, threading, and adjustments, correcting movements on the fly. This tested not only visual positioning and hand-eye coordination but also model inference, end-effector control, and flexible manipulation—truly realizing the concept that "trial and error is learning, and interaction is evolution."
Full-Stack Technology Foundation Unveiled
Bridging model capabilities to the real world requires synchronized support from data, execution hardware, and computing platforms. To that end, Anyverse Dynamics displayed its proprietary products: AnySense EGO, Anyverse ADA, and Anyverse ASC.
AnySense EGO is Anyverse Dynamics' proprietary first-person multimodal data acquisition system, comprising head-mounted and wrist-mounted terminals. It features 360° panoramic and detailed hand perception capabilities, fusing RGB with TOF depth information while supporting extensions for various end-effector collection modes. The system achieves microsecond-level synchronization of multi-source data like video and IMU, along with rapid cross-sensor coordinate alignment. With Wi-Fi real-time transmission and an ergonomic design, it meets the demands of long-duration continuous wear.
Anyverse ADA employs a 7-DOF anthropomorphic architecture, boasting a 30-millimeter wrist flip radius and a 606-millimeter working radius—ideal for dexterous operations in cramped, complex spaces. Weighing 7 kilograms with a rated load of 5 kilograms, the arm uses full-joint force sensing and real-time feedback to achieve compliant, safe human-robot and environmental interaction. It also supports desktop-level operations and high-fidelity skill migration.
Catering to varying scales of embodied intelligence computing needs, the Anyverse ASC Pro, Max, and Ultra form the complete lineup of Anyverse Dynamics' proprietary central computing platform. Covering a compute range from 600 INT8 TOPS to 4,000 FP4 TFLOPS, the platform is compatible with mainstream AI development frameworks and supports visual perception, intelligent decision-making, motion control, and end-to-end large model deployment.
The platform utilizes a multi-core heterogeneous architecture and dual SoC redundancy design, offering native support for Shadow Mode and over-the-air (OTA) updates. It also meets ASIL-B/(D) (ISO 26262) functional safety requirements.
Anyverse Dynamics is actively advancing the implementation of these solutions in relevant scenarios, having already achieved significant results in multi-scenario deployments with cumulative orders totaling 700 million yuan. The first batch of global deliveries has now begun. (Images courtesy of Anyverse Dynamics)
