For four decades, telling a machine what to do has meant one of three things: tapping a screen, pressing a button, or talking to it. All three assume you can stop what you’re doing, look down, and translate your intent into a tidy command. But a wind-turbine technician with a wrench in both hands, a dock worker in gloves, or someone steering an assistive mobility device on a crowded street doesn’t have that luxury. The robots have gotten remarkably capable. The way we talk to them hasn’t.
Wetour Robotics (NASDAQ: WETO), based in Austin, Texas, wants to flip that equation with Orchestra, a Physical AI operating system it unveiled on May 28, 2026. The pitch is deceptively simple: your body is the interface.
Rather than making the machine smarter, Orchestra makes the human a first-class node in the computing loop. It does this through what Wetour calls Spatial Intent Fusion — the simultaneous processing of three streams of human information: where your body sits in space, what your eyes are attending to, and what your muscles are about to do. Read in isolation, each channel is ambiguous. Fused in real time, they reconstruct intent.
The platform breaks into layers. VisionLink handles visual and spatial perception, feeding camera data into vision models that identify objects and estimate distances. Conductor is the biosignal pipeline, reading raw surface electromyographic (sEMG) data from a wrist-worn device and classifying it into gestures or continuous control signals. The clever part: sEMG signals appear at the skin roughly 50 to 80 milliseconds before a finger actually completes a gesture. Wetour calls this pre-motion intent sensing — the system anticipates rather than reacts.
All of this runs at the edge. The Orchestra hub packs 67 TOPS of compute and executes its full stack of vision, language, and neuromuscular models entirely on-device, with no cloud dependency on the critical path. Full-chain latency, from biosignal to actuator command, is held under 100 milliseconds — the threshold below which closed-loop control feels natural rather than laggy.
Wetour is refreshingly candid about where the seams show. Continuous sEMG recognition is reliable when you’re standing still, but walking or climbing introduces motion artifacts and electrode drift, so Orchestra falls back to a smaller set of robust discrete gestures in complex environments. Third-party actuators remain a fragmented mess of protocols, which Orchestra papers over with an AI-agent layer that negotiates connections adaptively.
The company frames this not as a rival to humanoid-robot developers but as their missing complement. Every natural human interaction with the physical world is a potential training signal — and most are currently invisible to any computer. Turn humans into observable nodes, the argument goes, and you generate the grounded, in-the-wild data that embodied AI badly needs.
An Orchestra Early Access program is open to startups, offering hardware, SDKs, and co-development support at no cost. Consumer pricing hasn’t been announced.
