The combination of artificial intelligence and robotics is considered a major step forward on the path to more powerful and flexible automation solutions. In this context, Agile Robots is increasingly focusing on Physical AI, in which intelligent software models interact directly with physical robot systems. A key component of this strategy is the use of the new Cosmos 3 platform, which combines various AI and simulation functions in a unified environment.

Cosmos 3 combines the generation of synthetic worlds, visual reasoning, and the simulation of actions within a so-called World Foundation Model. The goal is to create virtual environments in which robot systems can be trained, tested, and further developed in a realistic manner. The AI team at Agile Robots is currently exploring the platform’s capabilities, focusing in particular on simulation applications.

The integration of Cosmos 3 builds on a long-standing collaboration with NVIDIA. The company already uses various technologies from the semiconductor and AI specialist to develop and operate robotics solutions. These include, among others, the Isaac Lab development framework for robot learning and Jetson platforms for AI applications running directly on the robot systems. Participation in the European Industrial AI Cloud is also one of the joint projects.

The new platform has already been tested with various robot systems from the company. These include both single-arm and dual-arm robots from different series. The research focuses on how realistic virtual environments can be used to develop and optimize robotic workflows.

According to the company, Cosmos 3 is being used as a neural simulator for this purpose. The system can generate virtual scenarios in which robot behavior is tested under various conditions. This enables the prototyping of new capabilities, the simulation of rare or critical situations, and the validation of control models. Virtual testing is intended to shorten development cycles and reduce the effort required for real-world testing.

Video