Astribot T1 Reaches North America at $18,000: How Cable-Driven Arms Work

Astribot says its T1 wheeled humanoid is on sale in the US from $18,000, with immediate delivery, according to the company's 29 September 2026 press release. The robot made its North American debut at IROS 2026 in Pittsburgh. Its standout feature is a cable-driven body, which the company says aims to combine dexterity, compliance and back drivability. This is a company claim: no independent test of the T1 has been published as of 12 October 2026.

What is the Astribot T1?

The T1 is a wheeled, two-armed humanoid platform aimed at researchers and developers. It does not walk. A torso with two arms sits on a wheeled chassis, and the hardware is modular: users can add different end effectors, computing modules and sensors, per the press release.

The company-stated specifications are about 1.55 metres tall, about 66 kg, 23 degrees of freedom (DoF) excluding end effectors, and up to 5 kg of payload per arm. In plain terms, it is built to reach, grasp and tidy things at human height rather than to move through buildings on legs.

On the software side, Astribot offers an SDK and API covering joint control, Cartesian motion, whole-body coordination and sensor data. It also sells a layer called Astribot OS, with reusable skills and a natural-language interface meant to generate robot applications from a one-line requirement. The AI models are the company's own Lumo family.

What is Astribot's thesis?

Astribot's bet is that the body and the AI should be designed together. The company calls this approach "Design for AI". Per the press release, it builds three layers as one system: the Lumo models, the Astribot OS and the T1 body. It says it does not treat AI as software bolted onto an existing robot.

The press release also says the body is part of how the AI senses the world. A cable-driven system, it argues, can pass richer physical interaction information to the intelligence stack. That is a design argument, and the release offers no data to back it.

The company's learning loop is deliberately modest. Astribot says the T1 does not learn autonomously in real time during deployment. Instead it collects operation and teleoperation data, retrains models such as Lumo-2, and ships updated skills over the air or by on-site installation.

How is Astribot funded?

Gasgoo reported on 3 June 2026 that Astribot closed a Series B of more than 1 billion yuan, which it said pushes the valuation past 10 billion yuan. The article names no lead investor.

Listed participants are the Liangxi Sci-Tech Innovation Phase II Parent Fund (managed by Bohua Capital), Yangzhou Longtou Chiplet, Zhongbo Juli, ThunderSoft, Kede Education, an unnamed "major listed company" and Guoke Investment. Gasgoo describes Astribot as a cable-driven AI robotics company founded in 2022. These figures come from one report, not from a filing.

What are the Astribot T1's specifications?

Astribot has published a short list for the T1. Several items buyers would ask about are not published. All figures below are company-stated; none is independently measured.

How do cable-driven arms work?

A cable-driven arm moves its joints with cables or tendons, much like muscles pull on bones, instead of placing a geared motor inside every joint. The motors can sit near the base or in the torso. Cables then carry force out to the arm.

Research literature describes why that is attractive. A 2025 conference paper on a cable-transmission arm actuator says the approach lets an arm offload motor weight to an external structure, which raises payload capacity and lowers the arm's inertia. An April 2026 arXiv paper on a tendon-driven hand says relocating motors to the base or an external hub frees the moving parts for more DoF, sensing and maintainable structures.

For a normal user, lighter moving parts mean an arm that can swing fast and is less likely to hurt something if it bumps into it. Astribot makes a related claim: its cable-drive transmission, Gasgoo reports, mimics human tendon movement and force control, which it credits with dexterity, expressiveness and interaction safety.

Astribot has not said which T1 joints use cables, what the cables are made of or how they are tensioned. The press release gives no cable-specific numbers.

What are the downsides of cable-driven arms?

Cable drives trade one set of problems for another. The same literature names the costs: backlash, friction and length change.

The cable-transmission paper calls "considerable backlash" a challenge of Bowden-cable transmissions. Its authors respond with a tension-regulating mechanism and a recurrent-neural-network controller, and report a minimum absolute mean error of 2.3 degrees in simulation. That work targets surgical and rehabilitation uses, and the result is from simulation, so it says little about the T1.

The arXiv hand paper analyses tendon-sheath length variation and friction loss, and these shape how cables are routed and how joints are controlled. It reports 25 N of fingertip force through a 1 m remote sheath.

These papers study different robots from the T1, and neither says Astribot uses sheaths. They show what any cable-driven design must manage: cables stretch, rub and need tension control, so the maker must hold precision through software and mechanical design. Astribot's S1 sibling claims ±0.1 mm repeatability on its official product page, but that is a company figure and the page does not describe the cable mechanism.

How does the T1 compare with similar robots?

The closest like-for-like are research and developer robots that pair two arms with a body. Not all are wheeled and not all publish prices, so the table notes what is missing. Specs are company-stated, as of 12 October 2026.

Technology differences in short:

• Astribot T1 and S1. The same cable-drive idea at two scales. The S1 page lists 80 kg and the T1 release lists 66 kg.
• Unitree G1. A walking biped with conventional motor-driven joints. It is lighter and cheaper but lists a smaller arm load.
• Galbot G1. A wheeled robot with an NVIDIA AGX Orin 64GB compute module and a lifting waist and legs. It lists 5 kg per arm, but no price or actuator type.
• AgiBot A2-W. A heavier industrial wheeled platform with 7 DoF per arm. Price and actuator type are not published.

Where will people actually use the T1?

Astribot targets developers and researchers first. Practical uses are:

1. Research labs. Testing manipulation models on a body with an SDK for joint and Cartesian control.
2. Teleoperation data collection. The T1 supports VR-based teleoperation, which feeds model retraining.
3. Tidying demos. The company shows it gathering loose items into a backpack, including deformable objects.
4. App prototyping. Astribot says 15 teams built more than ten T1 applications in 36 hours at its second OS hackathon in Beijing, ending 21 September 2026.

What's still unclear?

Several points are open, and the cable-drive claim is the weakest.

• Price. The US price is "starting at $18,000" and the company does not say what that includes. The China price is 89,900 yuan (as relayed by Gasgoo), and the difference is unexplained.
• Cable design. Which joints use cables, how they are tensioned and how backlash is controlled are not published.
• Missing specs. Compute, sensors, battery capacity, runtime and speed are not published.
• Evidence. The backpack demo came without a success rate, and complex household tasks need more training, per the company.

Should you buy it or wait?

Buy only if you are a lab or developer who needs a wheeled, two-armed platform with a published SDK and can accept unverified claims. Wait if you want a home robot: Astribot itself says the T1 does not learn on its own in deployment and needs more training for complex household tasks.

Before ordering, ask Astribot for the exact $18,000 configuration, warranty and support terms, the compute and sensor list, and repeatability data for the T1 itself.