SenseTime’s ACE Robotics Unveils Three Core Technologies to Accelerate Embodied AI Deployment
Want to read in a language you're more familiar with?
SenseTime-backed ACE Robotics has introduced a new end-to-end technology stack aimed at turning embodied intelligence from research into real-world applications.
On December 18, 2025, ACE Robotics, a robotics subsidiary of SenseTime, officially announced three major technological breakthroughs: the ACE Embodied R&D Paradigm, Kairos World Model 3.0, the first open-source world model designed for commercial use, and the A1 Embodied Super Brain Module, injecting fresh momentum into the embodied AI industry.
The ACE Embodied R&D Paradigm follows a human-centric philosophy and uses environment-based data collection as its core engine, enabling the collection of tens of millions of data hours annually. This approach establishes a full-stack technical pipeline—environment-based data collection → Kairos World Model 3.0 → embodied interaction. By integrating multi-dimensional data, the system provides comprehensive “human–object–scene” coverage, offering advantages such as rich data dimensions, long task horizons, high interaction precision, and efficient data acquisition.
Kairos World Model 3.0 adopts an integrated multimodal understanding–generation–prediction architecture. It deeply models physical laws and human behavioral logic, enabling the generation of long-horizon dynamic interaction scenarios. The model supports 115 industry-specific scenarios, allowing developers to create task simulations through simple prompts. It has already been adapted to multiple China-made AI chips, including those from Moore Threads, Biren Technology, and Sugon.
The A1 Embodied Super Brain Module features cloud-based interaction capabilities, enabling it to interpret natural language and visual semantics and convert them into executable commands. This allows robotic platforms such as robot dogs to perform inspection, obstacle avoidance, and other tasks in complex environments, while accurately responding to user instructions.
