Part 4/8:
A fundamental challenge for robots lies in their ability to manipulate objects—a task humans often perform without conscious thought, honed over years of experiential learning. In contrast, robots generally lack this depth of experience, complicating their ability to adapt to new manipulation tasks. In response, researchers at Boston Dynamics are utilizing rich datasets from ongoing robotic operations—whether from real-world applications or simulations—to train AI models. This training aims to empower robots with enhanced abilities to generalize across diverse scenarios and self-correct during manipulation tasks, a significant shift expected to broaden the horizons of robotic capabilities in the coming years.