Astribot has brought its T1 humanoid research platform to North America at IROS 2026 in Pittsburgh, putting a surprisingly aggressive price on a robot built around dexterous manipulation: approximately ₿0.216 ($18,000) to start. The company is pairing the cable-driven machine with its Lumo embodied-AI models, developer interfaces and VR teleoperation tools rather than treating the robot as a closed appliance.
The new U.S. availability matters because T1 is not simply another walking humanoid demonstration. According to Astribot’s own platform description, the company designs the robot and AI stack together under a “Design for AI” architecture. Independent coverage of the IROS debut reports that T1 stands about 1.55 meters tall, weighs roughly 66 kilograms, provides 23 degrees of freedom excluding end effectors, and can carry up to 5 kilograms per arm.
Cable drive is the interesting hardware choice
Most of the attention around humanoids goes to foundation models, but T1’s cable-driven mechanics are arguably the more unusual engineering decision. Cable transmission can move motors away from the distal parts of an arm, reducing moving mass while preserving high force transparency and backdrivability. That combination is useful when a robot must make contact with glassware, deformable bags, tools or a human workspace instead of merely moving between collision-free poses.
A detailed IROS video and field report shows the robot performing manipulation demonstrations and highlights the 23-DoF architecture, 5-kilogram-per-arm payload and developer access. The same report notes that U.S. orders are open with immediate delivery advertised.
Lumo-2 tries to predict physics before generating actions
The software side is also more technically specific than the usual “AI-powered robot” label. Astribot describes Lumo-2 as a latent world-action model that predicts action-relevant physical changes in a compact latent space rather than generating dense future video frames. Its training pipeline progressively aligns actions with physical dynamics, then vision and language, before co-training across VLM, video and robot data.
That architecture is aimed at the awkward middle ground between understanding a command and surviving the physical sequence required to complete it. Astribot says Lumo-2 uses historical action context for long-horizon tasks and can train across robot demonstrations, general vision-language data and egocentric human video. The company reports a 2.71× end-to-end inference speedup over a standard autoregressive approach, although those are company-reported results rather than an independent benchmark.
At IROS, the practical demonstration was backpack packing. That sounds mundane until the object being manipulated is a soft bag whose geometry changes every time another item is inserted. The robot has to keep track of a changing opening, moving fabric, object placement and the sequence of actions rather than replaying one rigid trajectory.
A research robot priced closer to a workstation
At roughly ₿0.216 ($18,000), T1 enters a price range where university labs, robotics startups and small research teams can at least consider owning a full platform instead of negotiating access to a six-figure machine. The number is a starting price, not a claim that every sensor, hand, compute module or deployment configuration is included.
The platform also supports modular end effectors, computing modules and sensors, plus SDK/API access for joint control, Cartesian motion, whole-body coordination and sensor data. That developer-first positioning puts T1 in the same broader shift BitcoinVersus.Tech has tracked with Feather’s developer robot and the increasingly open tactile hardware shown by Sharpa.
There is an important limitation: T1 is still a research and development platform. A polished conference demo does not establish reliability across thousands of unscripted household or industrial cycles. Astribot’s workflow also retrains models from collected robot and teleoperation data; the robot is not autonomously teaching itself new jobs in real time.
Why the architecture matters
The interesting bet is that manipulation quality may become a stronger differentiator than humanoid appearance. A robot that can predict physical state changes, feel compliant contact through a low-inertia mechanism and expose its control stack to researchers can be useful even without legs. That is a different path from the warehouse emphasis behind Agility’s Digit 5.
If T1’s price, immediate availability and developer access hold up outside the show floor, the platform could make embodied-AI experimentation materially cheaper. The bigger test now is repeatability: whether the same cable-driven hardware and Lumo stack can move from backpack and lab demos into long-duration work where failures, calibration drift and messy environments expose what a conference demonstration cannot.
BitcoinVersus.Tech
Advertisement
BitcoinVersus.Tech Editor’s Note:
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiatives, please donate here: 3C9o19EH5HSiwEPyCTmEKzxhNCbo2X6TTb
BitcoinVersus.tech is not a financial advisor. This media platform reports on financial subjects purely for informational purposes.
Leave a comment