Robot Awakening Cognitive Robotic Capabilities for New Opportunities in Manufacturing

Related Vendors

Fraunhofer-Institut für Fabrikbetrieb und -automatisierung IFF

FAULHABER Drive Systems

dataTec AG

Wuxi InfiMotion Technology Co., Ltd.

Researchers have given robots new abilities that enable them to take on tasks that were previously not automatable. In addition, there are new safety systems and planning tools.

With new AI-based (AI = artificial intelligence) developments, researchers at the Fraunhofer Institute for Factory Operation and Automation (IFF) are equipping robots with the necessary cognitive abilities to autonomously operate in unstructured, changing environments. This means they could automatically perform complex processes such as assembly and disassembly in industrial settings or object handling in the care sector even when these situations arise. Projection and camera-based safety technology enables robots with AI-based motion control to reliably respond to changes, adapt to new tasks, and finally complete the application safely, as further stated.

Cognitive Robots Learn from Experience

This opens up a broad spectrum of new application fields that were previously closed to conventional robotics, which were limited to specific, narrowly defined tasks. Because cognitive robots can learn from experiences, make independent decisions, and adapt to various scenarios, as emphasized by IFF. For "Pick & Place" tasks, for example, which involve picking up and placing components, a cognitive robot no longer has to learn what the individual workpieces look like before it can grasp them. Instead, it uses its camera to capture the size, shape, texture, and condition of the object. With this information, it can adjust its behavior accordingly, allowing it to handle different environmental conditions and even different packaging materials.

Environment Simulation Needs to be Learned

To train the AI models used, the experts utilize simulation environments. For example, when focusing on assembly and disassembly processes, such as the removal of motherboards from a PC, numerous virtual robots can be trained simultaneously and at a much higher speed without safety concerns in the digital space. Learning in digital simulation has many advantages, but also a weakness! The virtual learning environment is never 100 percent identical to reality. Therefore, researchers must manage to close the reality gap—also known as the "Sim2Real" gap—as much as possible. The goal is to make the simulation either as identical to reality as possible or to cover as many real-world variations as possible so that the neural network used for AI learns to generalize and navigate unfamiliar environments, as explained. This is achieved, among other things, through domain randomization. With it, a variety of simulated environments with random properties can be created, and a model that works in all environments can be trained, it is stated. Different lighting conditions, for example, influence the simulation. However, this set of parameters can be changed during training. The robot thus learns not to solve the exact simulation but to understand the abstract concept behind it. Reality essentially becomes a new variant of a simulation for the AI.