Physical AI
AI applied to systems that sense and act in the physical world. A robot must account for its environment, equipment and operating constraints as well as its assigned task.
Research & development
Physical AI and foundation models
We aim to connect robots’ understanding of their surroundings with task execution through a reusable software and learning foundation. Our physical AI work builds on integration experience and advances through defined research and validation tasks.
Foundation model
A model trained on broad data that can be adapted to different tasks. For robotics, our research direction is to bring visual observations, sensor information and task descriptions into a shared learning approach.
Different robots, shared learning
Cross-embodiment describes the ambition to reuse learning across different robot bodies. Each robot still requires its own interfaces, adaptation and validation; a shared model does not establish universal compatibility.
Simulation and real-world validation
Digital environments allow repeatable exploration of scenarios before physical trials. Sim-to-real work examines how well those results transfer to actual equipment and conditions.
Our shared technology approach
One Brain
Shared task software, robot adapters and reusable skill libraries.
Quad Foundation Model
Research into learning from observations, task descriptions and robot state.
Cloud Robotics
Development of shared task records, monitoring and coordination across compatible systems.
Application research
Production and service robotics
Research into task understanding and reusable skills for handling parts, interacting with equipment and coordinating service workflows.
Mobile inspection
Research into relating observations to inspection tasks and making collected information useful to operators.
How we intend to measure progress
We plan to evaluate bounded tasks in simulation and physical trials, considering task completion, operator intervention, repeatability and transfer to another compatible platform. Data quality, permitted data use, model licensing and version tracking are part of this development approach.
This is an R&D direction. Platform coverage, deployment scope and demonstrated performance will be established through project-specific validation; it is not a claim that one model already operates every robot.