Shenzhen-based Physical AI company Astribot is making its North American debut at IROS 2026, the International Conference on Intelligent Robots and Systems, running September 27 through October 1 at the David L. Lawrence Convention Center in Pittsburgh, Pennsylvania. The company is showcasing its T1 humanoid robot alongside its full Physical AI stack, marking the first time Astribot has brought its integrated platform to a North American audience.
One System, Not Three
What sets Astribot apart is its “Design for AI” philosophy, which treats the company’s AI models, embodied operating system and cable-driven robotic body as a single, co-designed system rather than separate components bolted together after the fact.
The idea is straightforward: intelligence, software and hardware should be built around each other from the ground up. It’s a meaningful distinction in a field where AI is often treated as a software layer dropped onto an existing robot frame.
The Lumo Model Family
At the heart of Astribot’s approach is the Lumo model family, built to connect reasoning with physical action. Lumo-1 introduced the company’s Reasoning-Action Foundation Model framework, while Lumo-2 explores latent world-action modeling for handling more complex physical tasks.
Running on Lumo-2, T1 has already demonstrated autonomous tidying, including sorting miscellaneous items into a backpack, a task that requires both decision-making and dexterous handling of deformable objects. Astribot continues to develop Lumo-2 for broader household applications, with additional capabilities dependent on further training.
An Operating System Built for Real-World Use
A capable AI model is only part of the equation. Astribot’s embodied operating system bridges AI capabilities with people, robots and physical environments. For developers, T1 offers SDK and API access covering joint control, Cartesian motion, whole-body coordination and sensor data.
The OS also includes meta-packages and skill libraries for orchestrating agentic behaviors, along with a natural-language interface that can generate deployable robot applications from a single-line requirement.
At the second Astribot OS Hackathon, held in Beijing and concluding September 21, 15 teams built and demonstrated more than ten functional T1 applications in just 36 hours.
Built to Move and Adapt
T1 stands approximately 1.55 meters tall, weighs around 66 kilograms and features 23 degrees of freedom (excluding end effectors), with a payload capacity of up to 5 kilograms per arm. Its cable-driven architecture supports dexterous manipulation, compliant interaction and high-speed movement, while also feeding richer physical interaction data back into the AI stack.
Practical considerations are covered as well: T1 supports automatic charging and quick battery replacement, and its modular design allows different end effectors, computing modules and sensors to be swapped in for different tasks.
How T1 Learns Over Time
T1 does not learn autonomously in real time during deployment. Instead, Astribot collects multimodal data from robot operation and teleoperation, curates it, retrains models like Lumo-2 and pushes updated skills via over-the-air updates or on-site installation. Customers can opt to contribute data from autonomous task execution or dedicated training sessions, feeding back into the broader development cycle.
Pricing and Availability
T1 is positioned as a developer-accessible Physical AI platform for researchers, developers and real-world application builders. US pricing starts at $18,000, with orders open now and immediate delivery available. The platform includes modular hardware, developer interfaces, VR-based teleoperation and AI capabilities out of the box. More details on the T1 platform and Astribot’s Physical AI approach can be found at astribot.com.
