Archon Robotics, a Shanghai startup, has published its first public demo: a three-minute video introducing “Whole-Body Intelligence” (WBI), its name for AI that connects task understanding to the coordinated movement of an entire legged humanoid. The debut, shared on X on October 12, shows modified Unitree G1 humanoids, Booster T2s and an AGIBOT A3 working through kitchen and dining tasks — clearing tables, loading a dishwasher, organising a refrigerator, mopping — with one wide scene fitting three modified G1s and two T2s into a single shared space, each on a different job. The company’s tagline: “Building Whole-Body Intelligence for the Real World. We’re Just Getting Started.”
The name should sound familiar. When Google DeepMind launched Gemini Robotics 2 in July, it framed the release around the same phrase — whole-body intelligence for humanoid robots, the model controlling everything from feet to fingertips. Four months later, a Shanghai lab has adopted the identical term as its product category. Whether that is deliberate positioning or convergent vocabulary, it marks out the turf Archon wants: the software layer that choreographs balance, locomotion and manipulation as one problem rather than three.
Inside the video, the system is presented as a two-part split. S1 handles task understanding and multimodal information; S0 handles coordination, stabilisation and precise execution. Archon calls S0 the robot’s “cerebellum” — the layer that translates targets into executable whole-body movement. Pretraining draws on human data, with post-training done on real robots. A July technical blog had described a four-layer framework, adding task planning and motion generation on top; the debut shows only S1 and S0, and as Humanoids Daily notes, it is unclear whether that reflects an architectural change or simply a narrower presentation. Asked to compare, there is nothing to compare: the company says it has tested the approach “on humanoid platforms with different configurations” but provides no comparative results in the video.
One detail deserves more attention than debut reels usually get. Archon’s closing disclosure separates its footage types: the full demonstrations run autonomously at normal speed, while the S0 explanations and the DAgger segments — the corrective-data collection that sharpens the policy after mistakes — show teleoperation. That is a cleaner distinction than most first videos bother to draw, and it matters for reading the robot’s movements: what looks effortless and what was hand-held are at least labelled.
As of publication, the debut has not been independently reported outside Humanoids Daily’s write-up of the company’s post; the story rests on Archon’s own footage and claims.
There are two ways to read what Archon is doing. The first is the obvious one: another entrant in an increasingly crowded field of physical-AI startups betting that intelligence, not hardware, is the scarce resource. Like MindOn and Gemini Robotics 2, Archon is deliberately platform-agnostic — it has no robot of its own to sell, running its stack on machines from AGIBOT, Booster and Unitree. That bet only pays if the software genuinely transfers between bodies, which is precisely the claim the demo gestures at (“different configurations”) without substantiating.
The second reading is about vocabulary. Whoever names the category frames every comparison that follows — if “whole-body intelligence” becomes the standard label for full-body manipulation models, the company that co-owns the phrase gets cited in every write-up, including this one. The physical-AI race has already produced one contest over what counts as evidence (DeepMind versus OpenAI on what demos actually prove); naming is the quieter sibling of that fight. Archon’s demo is competent-looking but early — no benchmarks, no numbers, no timeline. The honest verdict on a debut video is the one the company supplies in its own tagline: just getting started.
Source note: story draws on Humanoids Daily’s report, Archon’s X post and the company’s July technical blog; the Korthos Robotics directory profiles the company.
