Las Vegas, first week of January 2026. Thirty-eight humanoid robots invade the CES — including twenty-one Chinese manufacturers, representing 55% of the category. The images that have gone viral on social media show two Unitree G1s boxing in a BattleBots arena, the H2 performing consecutive backflips and kicks, the G1 dancing in synchronized choreography, and the R1 in kung-fu postures. Much concluded it was spectacular folklore. This diagnosis is insufficient — and probably contrary to what was actually at stake that week.
The teleoperated spectacle, a new commercial funnel
What media coverage has largely overlooked: the Unitree H2 integrates a teleoperation capability via portable rigs and the Apple Vision Pro. This is not a technical specification detail. Unitree’s public repository xr_teleoperate explicitly describes the tool as a data collection toolkit for embodied intelligence, designed to record the gestures of the human operator — Vision Pro, Quest, or PICO — in order to feed imitation learning. The spectacular gesture and data capture are the same gesture.
Analysts present at the CES confirmed that the vast majority of the systems on display were at autonomy level 0 or 1 — scripted or teleoperated. Boston Dynamics' Atlas demonstration at the show was also remotely controlled. This fact, rather than invalidating the robots, reveals the true architecture of the model: teleoperation is no longer an admission of weakness to be hidden, it is the data collection mechanism that the sector has industrialised.
The slashed price as a data acquisition cost
The Unitree R1 is offered from 4,900 USD, the R1 Air version. It is positioned as a developer-first platform. Reading this price as competitive dumping would be a framing error. Putting a teleoperated body in the hands of thousands of laboratories, integrators, and early adopters at this price is massively distributing collection sensors: every demo, every teleoperation session feeds the imitation learning dataset. The R1 at 4,900 USD is not sold at a loss to kill hardware competition — it is sold to accelerate the accumulation of movement data.
Unitree has also announced at CES 2026 an explicit pivot towards a Robot-as-a-Service model, with a skills marketplace. The intention is clear: the robot is the terminal; the value lies in the layer of data and skills that accumulates above it.
NEURA, LG, AgiBot: three data taps
The same movement can be seen among other major players at the show, each through their own path.
NEURA Robotics presented the Neuraverse: an "invisible" operating system that connects robots and networked devices to share collective learning. The integrated marketplace allows developers to publish, share, and monetise robotic skills. NEURA robots are trained in NEURA Gyms — real physical environments, not just simulation — to generate physical AI data. The training infrastructure is the product as much as the robot.
LG Electronics unveiled the CLOiD, a domestic robot whose VLA (Vision Language Action) model was trained on tens of thousands of hours of household task data. Simultaneously, LG took stakes in Figure AI (United States), AgiBot (China), and Dyna Robotics. This investment portfolio is not a geographical hedge in the classic financial sense. AgiBot is precisely known for its data learning farms — collection and training infrastructures on movement data. LG is not buying diversified hardware bets: it is securing data taps on two continents.
What was autonomous, and what was not filmed
The only autonomy truly demonstrated at CES 2026 went almost unnoticed, precisely because it is boring to watch. X-Humanoid (Beijing Humanoid Robot Innovation Center) exhibited two machines: the Tien Kung 2.0, which performs industrial parts sorting in full autonomy, and the Tien Kung Ultra, which ran a 21.0975 km semi-marathon without a remote control in 2h40m42s. No choreography, no arena. These two performances did not go viral.
The practical demonstrations were nevertheless well present: preparing drinks, folding laundry, distributing cards, sorting parts. But it was the backflips and boxing matches that structured the coverage. The hierarchy of attention is itself a revealer: the spectacular teleoperated captures the eye and the share; the useful autonomous is invisible. The sector has integrated this filter and turned it into an advantage.
The signal from EngineAI’s CEO
An incident on the sidelines of the CES illustrates the latent tension. Zhao Tongyang, founding CEO of EngineAI, published a video showing him kicking the T800, filmed from multiple angles in a studio, to respond to doubts about the authenticity of his robot’s capabilities. The fact that a leader must provide multi-angle proof to convince that his robot moves on its own says something about the state of market trust — and the growing pressure to distinguish what is autonomous from what is staged.
The shift, not the revolution
In December 2025, this chain was already showing that choreography is a product sheet, that the R1 functions as a data probe, and that manufacturers had to prove on video that a robot moves without help. CES 2026 is the next step in this same arc: the sector is no longer ashamed of teleoperation, it has transformed it into an economic model. The dance is no longer a lie to hide — it is the customer acquisition cost of a data platform that Unitree, NEURA, LG, and AgiBot are now openly competing for.
Tracking the number of robots at a show, or comparing their acrobatics, is looking at the terminal. The battle is played on the layer above: who accumulates the richest movement data, who controls the skills marketplace, who owns the training infrastructure. Las Vegas was the setting. The product was the collection.