Astribot, a Shenzhen, China-based bodily AI firm, showcased its T1 humanoid robotic to North America for the primary time at IROS 2026, which occurred from September 27 to October 1 on the David L. Lawrence Conference Middle in Pittsburgh, Pennsylvania.
The corporate demonstrated its built-in strategy to bodily AI spanning repeatedly bettering AI fashions, an embodied working system and a cable-driven robotic physique.
Somewhat than treating AI as a software program layer added to an current robotic, Astribot follows a “Design for AI” philosophy, designing intelligence, the working system and the bodily physique as an built-in system.
AI: Intelligence designed to behave
Astribot says it locations AI on the heart of its bodily AI structure. Its Lumo mannequin household is designed to attach reasoning and motion in bodily environments, with the corporate creating fashions that may understand, cause about and execute duties by means of real-world robots.
Lumo-1 launched Astribot’s Reasoning-Motion Basis Mannequin strategy, whereas the corporate’s subsequent Lumo-2 work explores latent world-action modeling for extra advanced bodily duties.
The aim isn’t merely to provide a stronger mannequin in isolation, however to repeatedly translate advances in AI into new capabilities for bodily robots.
Geared up with the Lumo-2 base mannequin, T1 has demonstrated autonomous tidying of miscellaneous objects right into a backpack, combining decision-making with dexterous manipulation of deformable objects.
Astribot is creating Lumo-2 to increase these capabilities to extra advanced family duties; extra process capabilities rely on additional coaching.
Embodied OS: Connecting AI to the bodily world
An AI mannequin alone doesn’t make a robotic helpful. Astribot is subsequently creating an embodied working system designed to attach AI capabilities with individuals, robots and bodily environments whereas decreasing the obstacles to interplay, improvement and use.
For builders, T1 gives SDK and API entry to capabilities together with joint management, Cartesian movement and whole-body coordination, in addition to entry to sensor knowledge.
Astribot OS gives meta-packages and talent libraries that assist builders orchestrate agentic behaviors and foundational features. Its natural-language interface is designed to generate deployable robotic purposes from a single-line requirement.
On the second Astribot OS Hackathon, which concluded September 21 in Beijing, 15 groups developed and demonstrated greater than ten purposeful purposes on T1 in 36 hours, in keeping with the corporate.
Robotic Physique: Designing the bodily layer for AI
The third layer is the robotic physique itself. T1 is constructed round Astribot’s cable-driven structure, which is designed to assist dexterous manipulation, compliant interplay and high-speed motion.
The robotic stands roughly 1.55 meters tall, weighs about 66 kilograms, has 23 levels of freedom excluding finish effectors, and helps as much as 5 kilograms of payload per arm.
T1 additionally helps automated charging and fast battery substitute, whereas its modular design permits totally different finish effectors, computing modules and sensors to be added for various duties.
Astribot’s “Design for AI” strategy displays the corporate’s view that the bodily physique needs to be designed round what AI must understand and do, relatively than optimized independently from intelligence.
In a cable-driven system, the corporate goals to mix dexterity, compliance and again drivability with the flexibility to show richer bodily interplay data to the intelligence stack.
This makes the physique greater than an actuator platform: it turns into a part of the system by means of which AI interacts with the bodily world.
Studying by doing
The three layers type a steady improvement loop: AI generates capabilities, the embodied OS connects these capabilities to the robotic and its customers, and the bodily physique executes them in the actual world. T1 doesn’t study autonomously in actual time throughout deployment.
As an alternative, Astribot collects multimodal knowledge from robotic operation and teleoperation, processes and curates that knowledge, retrains fashions reminiscent of Lumo-2, and deploys up to date abilities by means of over-the-air updates or on-site set up and calibration.
Prospects can select to gather knowledge for devoted mannequin coaching or throughout autonomous process execution. For Astribot, that is the inspiration of its strategy to bodily AI: AI that grows by doing.
T1: Making bodily AI extra accessible
T1 is designed as a developer-accessible bodily AI platform for researchers, builders and real-world purposes. With US pricing beginning at $18,000, T1 combines modular {hardware}, developer interfaces, VR-based teleoperation and AI capabilities. US clients can order T1 now, with speedy supply out there.
