Shanghai-based ACE ROBOTICS used a WAIC 2026 forum on Saturday to unveil Kairos 3.1, its latest action-oriented world model, alongside Ambient Capture Engine 2.0 and three commercial robot solutions spanning instant retail, hotel laundry and outdoor services. Chairman Wang Xiaogang, a SenseTime co-founder, framed the launch as physical AI's move from "understanding" to "execution."
A single model for physical reasoning
Kairos 3.1 is a natively unified architecture that fuses generative, physical and cognitive intelligence in one hybrid Transformer. Visual observations, language instructions, force and tactile signals and policy trajectories share a latent space so a robot can plan, simulate physical cause and effect, execute and reflect inside a single loop.
At the core is ACE-BRAIN-0.5, which ACE ROBOTICS says posted state-of-the-art results on 12 public spatial-understanding, navigation, manipulation and task-progress benchmarks as of July 2026. Kairos-HomeWorld generates whole-home scenes trained on 300,000 residential floor plans and 8,700 3D assets. The 8-billion-parameter Kairos 3.1 runs at 125 ms inference latency on NVIDIA's Jetson Thor through the in-house KairosRT engine.
An Information-Density Law for embodied data
ACE ROBOTICS also introduced what it calls the Information-Density Law: data's value is measured not by volume but by whether it can change an agent's next action. Its five-level L1-L5 taxonomy culminates in L5 data that captures three-dimensional force, failure-recovery trajectories and open-environment variables. Built on that principle, Ambient Capture Engine 2.0 combines a wearable ACE Ego Kit (with a 0.01-newton-sensitive ACE Sense Glove), the automated ACE Data Engine and the ACE Ego Matrix standard. The company also open-sourced ACE-Data-0, a household-interaction dataset at the L5 level.
Three commercial deployments
The Kairos stack now anchors three named solutions. Xiaoman is an instant-retail platform paired with the new W1 fulfillment robot, live inside Sense MartGo, Kuaikeda and PetroChina convenience stores, with plans to scale to 1,000 sites within a year and roughly 10,000 within two. Xiaoxin targets hotel laundry, a category where ACE ROBOTICS says roughly 80% of demand hits at night. Xiaotu applies an "one brain, many bodies" architecture to quadruped platforms for outdoor navigation and visitor services, already deployed at the Begonia Flower Festival in Tianjin.
To back the training and serving requirements, ACE ROBOTICS launched a World Model Cloud Ecosystem initiative with Baidu AI Cloud, Alibaba Cloud, Huawei Cloud, Tencent Cloud and SenseCore AI Cloud. The launch dovetails with a WAIC that has already seen Shanghai-based rivals such as AgiBot unveil four new humanoids and Beijing peers push open-weight models on the same stage.
Reporting based on coverage from The Manila Times and Bastille Post.