What is physical AI? (www.ibm.com)

Physical AI is emerging as a groundbreaking concept within the AI/ML community, allowing AI systems to interact with and act upon the physical world through the integration of sensors, actuators, and AI models. Traditionally, robots relied on scripted behavior, limiting their tasks to specific environments. However, with advancements in large language models (LLMs) and reinforcement learning, robots can now leverage common sense reasoning and adapt to complex tasks in varied environments. This marks a significant shift toward more autonomous, intelligent robots capable of learning from their experiences, such as navigating crowded settings or performing intricate tasks like litter collection. The significance of physical AI is heightened by three converging trends: the rise of generative AI through foundation models, advancements in high-fidelity simulation for training purposes, and improvements in sensor technology and hardware. These innovations help overcome previous bottlenecks, enabling efficient model training and real-world application. Nvidia's CEO, Jensen Huang, predicts that physical AI could lead to a new global economy centered around a billion robots, potentially kickstarting a second industrial revolution. The release of Nvidia's open models and frameworks for physical AI further bolsters this development, providing essential tools for rapid robot development and fostering a renaissance in physical automation across industries.