At this year's Automate conference, the first eye-catching news came from Intrinsic, a company spun off from Alphabet X. During the event in Chicago on Monday, the company announced the integration of their Flowstate robot application platform with some of NVIDIA's products. One of the products involved is the Isaac Manipulator, a series of foundational models focused on building workflows for robot arms. These models made their debut at the GTC conference in March this year and have already attracted leading companies in the industrial automation field, such as Yaskawa, Solomon, PickNik Robotics, Ready Robotics, Franka Robotics, and Universal Robots. This collaboration specifically focuses on grasping technology, which is a crucial step in manufacturing and fulfillment automation. These systems are trained on large-scale datasets to achieve hardware-agnostic and task execution across different objects. In simple terms, this means that grasping technology can be transferred to different environments without the need to retrain each system for every new scenario. Just like humans, once we learn how to pick up things, this ability can adapt to different objects in different environments. However, currently, most robots are unable to do this, but Intrinsic is working towards this direction. Wendy Tan White, the founder and CEO of Intrinsic, stated in an article, "In the future, developers will be able to leverage these pre-trained general grasping skills to significantly speed up their programming process. For the entire industry, this progress means that foundational models could have far-reaching impacts, making robot programming challenges easier to manage at scale, creating applications that were previously unachievable, reducing development costs, and increasing end-user flexibility." Early testing of Flowstate was conducted on NVIDIA's robot simulation platform, Isaac Sim. One of Intrinsic's clients, Trumpf Machine Tools, has been closely collaborating with the system's prototype. "These general grasping skills are trained on 100% synthetic data in Isaac Sim and can be used to build complex solutions that perform adaptive and diversified object grasping tasks in simulation and reality," Tan White said when discussing Trumpf's collaboration with the platform. "Unlike hardcoding specific fixtures to grasp specific objects in a specific way, this foundational model generates efficient code for specific fixtures and objects to complete tasks." In addition, Intrinsic is also collaborating with DeepMind, another subsidiary of Alphabet, to break through the key areas of pose estimation and path planning in automation. Specifically, these systems have been trained on over 130,000 objects and can determine the orientation of objects in "a matter of seconds," which is a crucial prerequisite for picking up objects. Another highlight of Intrinsic's collaboration with DeepMind is the ability to coordinate multiple robots. Tan White stated, "Our team has tested this fully machine learning-generated solution to seamlessly coordinate four individual robots working in a scaled-down automotive welding application simulation. The motion plans and trajectories for each robot are automatically generated, collision-free, and remarkably efficient—about 25% better than some of the traditional methods we have tested." The team is also developing a dual-arm robot system, which is closer to the emerging world of humanoid robots. In the coming years, whether it's humanoid robots or not, we will see more of these dual-arm systems. The transition from one arm to two arms will open up new application areas for these systems.