A motion-first approach to smarter factory robots

The article discusses the limitations of traditional industrial robots in handling real-world variability during manufacturing tasks, particularly assembly. While robots excel in controlled, repetitive tasks like welding and painting, they struggle with assembly due to the high costs and inflexibility of programming and integration, which can account for up to 70% of total robot application expenses. Standard robots rely on a recognition-first approach using machine learning models that require extensive pre-training on large datasets to identify objects before interacting with them. This approach falters in dynamic environments where objects vary in appearance due to lighting, orientation, or transparency, such as glass bottles, making reliable recognition and handling difficult. An Indian startup, CynLr, proposes a novel “motion-first vision” system embodied in their CyRo robot, which learns by physically interacting with unfamiliar objects in real time rather than relying on pre-trained models. This approach bypasses the recognition bottleneck by focusing on how objects can be manipulated rather than solely identifying them visually. The system addresses fundamental engineering challenges