The first news of this year's Automate conference comes to us via Alphabet X spinout Intrinsic. The company announced at an event in Chicago on Monday that it will incorporate a number of Nvidia products into its Flowstate robot app platform.
This includes Isaac Manipulator, a collection of basic models designed to create robotic arm workflows. The product was announced at his GTC in March, and major companies in industrial automation are already on board. The list includes Yaskawa, Solomon, PickNik Robotics, Ready Robotics, Franka Robotics, and Universal Robots.
This collaboration has a particular focus on grasping (grasping and lifting objects), one of the key techniques for both manufacturing automation and fulfillment automation. The system is trained on large datasets with the aim of performing tasks that work across hardware (i.e., hardware-independent) and on a variety of objects.
This means you don't have to train every system for every scenario, and you can transfer picking methods to different settings. As humans, once we understand how to pick up things, we can adapt that behavior to different objects in different environments. Most of the time, robots can't do that. At least not for now.
Image credit: Unique
“In the future, developers will be able to use these off-the-shelf general purpose grasp skills to significantly accelerate the programming process,” wrote Wendy Tan White, Founder and CEO of Intrinsic. It is stated in. “For the broader industry, this development will help make today's robot programming challenges more manageable at scale, create previously unfeasible applications, reduce development costs, and increase end-user flexibility. It shows how the underlying model can have a huge impact.”
Initial Flowstate testing was conducted on Isaac Sim, Nvidia's robot simulation platform. Trumpf Machine Tools, an essential customer, is working on a prototype of the system.
“This universal grasping skill, trained on 100% synthetic data in Isaac Sim, can be used to build sophisticated solutions that can perform adaptive and versatile object grasping tasks in simulation and reality.” Tan White said of Trumpf's work on the platform. “Instead of hard-coding a specific gripper to grasp a specific object in a specific way, efficient code for specific grippers and objects is automatically generated and uses the underlying model to complete the task. ”
Intrinsic is also working with DeepMind, also part of Alphabet, to crack two important aspects of automation: pose estimation and path planning. In the latter case, the system was trained on over 130,000 objects. The company says the system can determine the orientation of an object in “seconds.” This is an important part of picking up objects.
Another key part of the Intrinsic and DeepMind collaboration is the ability to have multiple robots work together. “Our team tested this his 100% ML-generated solution to seamlessly coordinate his four separate robots working in a scaled-down automotive welding application simulation,” said Tan White. Masu. “The motion plans and trajectories for each robot are automatically generated, collision-free, and incredibly efficient, outperforming some traditional methods we've tested by up to 25%.”
The team is also working on a system that uses two arms at the same time. This is a setup more in line with the emerging world of humanoid robots. We're going to see a lot more of them in the coming years, humanoid or not. Moving from one arm to two arms opens up a world of further applications for these systems.