ACE Robotics Launches Kairos 3.1 World Model at WAIC 2026, Targeting Commercial Embodied AI in Retail and Hospitality

ACE Robotics announced Kairos 3.1, an action-oriented world model for embodied AI, at the 2026 World Artificial Intelligence Conference in Shanghai on July 19. The launch included Ambient Capture Engine 2.0, three commercial solution packages, and a new industry benchmark called Physical IQ, according to a company announcement distributed via Media OutReach Newswire.

Architecture and Capabilities

Kairos 3.1 integrates generative, physical, and cognitive intelligence within a single hybrid Transformer architecture. It uses a shared mixed-attention mechanism to bring visual observations, language instructions, force and tactile signals, and policy trajectories into a unified latent space, according to ACE Robotics.

The model supports what the company describes as an “understand, reason, execute, and reflect” loop. It breaks down long-horizon tasks, simulates physical cause and effect across multiple scenarios, ranks candidate actions, executes the selected strategy, and evaluates the outcome for adjustment.

At the core of its spatial understanding is ACE-BRAIN-0.5, which ACE Robotics said achieved state-of-the-art results across 12 public evaluations covering spatial understanding, navigation, manipulation, and task-progress assessment as of July 2026. In internal testing, the Kairos 3.1 8B model achieved inference latency of 125 milliseconds on an NVIDIA Jetson Thor platform at BF16 precision.

The model also incorporates self-reflective iteration. In one test scenario described in the announcement, a robot changed from a three-finger to a four-finger grasp after an unsuccessful attempt and subsequently completed the task.

Data Infrastructure: Ambient Capture Engine 2.0

Alongside the model, ACE Robotics introduced Ambient Capture Engine 2.0, a system for capturing, processing, and reusing high-density physical interaction data.

The hardware component, ACE Ego Kit, is a lightweight wireless wearable with head, hand, and chest modules. It includes the ACE Sense Glove, which offers sensitivity of 0.01 newtons and a joint-angle error of less than two degrees. The system synchronizes data from more than 20 heterogeneous sensors with a timing error of less than one millisecond.

The software component, ACE Data Engine, includes ACE-ViDiHand, a generative 4D hand-motion-capture framework that ACE Robotics said leads three public benchmarks (ARCTIC, HOT3D, and HOI4D) as of July 2026, with frame-level accuracy of 0.997 and up to 4.8x improvement in capture fidelity.

Commercial Deployment and Benchmark

ACE Robotics announced three commercial solutions targeting instant retail, hospitality service, and outdoor deployment scenarios. The company did not disclose pricing or initial deployment partners.

The Physical IQ benchmark was co-launched with the Shanghai Artificial Intelligence Association, the Shenzhen Loop Area Institute, and the East China branch of the China Academy of Information and Communications Technology, with participation from more than 20 universities and industry partners. It aims to provide a unified evaluation framework for embodied physical AI, filling a gap in standardized measurement for robotic manipulation and task execution.

Context for Agent Builders

The release extends the “agent” concept beyond text and reasoning into physical execution. Most agent infrastructure development in 2026 has focused on software agents: coding tools, workflow automation, and enterprise orchestration. ACE Robotics is positioning Kairos 3.1 as the equivalent planning substrate for physical tasks, where models must reason about force, spatial relationships, and failure recovery rather than token sequences. The WAIC 2026 forum was organized around advancing physical AI from “understanding” to “execution.”