Enhanced perceptionAdvances in sensors and AI have dramatically improved robots’ ability to perceive their surroundings. Affordable high-resolution cameras, light detection and ranging (LiDAR) and next-generation tactile sensors, among other sensors, give robots richer raw inputs, while advanced computer vision algorithms (powered by deep learning) enable visual perception approaching human-level capabilities. Robots can now recognize and interpret complex environments in real time – identifying objects, recognizing their 3D orientation and assessing their physical properties – essential prerequisites for developing an understanding of how to interact with objects. These advances allow robots to “see” and comprehend an object and its environment with unprecedented clarity. Autonomous decision-making and planningInnovations in AI and software have enabled robots to make intelligent decisions in real time. Instead of rigid pre-programming, robots now exploit reinforcement learning and simulation to learn behaviours through trial and error in virtual environments. Advanced simulators (e.g. high-fidelity physics simulators) and domain randomization techniques (e.g. randomization of parameters such as lighting or friction) are closing the simulation-to-reality gap, so that behaviours learned in simulation transfer seamlessly to real machines. Robots also increasingly benefit from powerful foundation models that integrate vision, language and action. These models, such as Google DeepMind’s Gemini Robotics6 and Nvidia’s Isaac GR00T,7 ingest multimodal inputs and generate task-appropriate outputs – allowing for intuitive human–robot interactions and superior contextual understanding. This enables robust workflow planning: given a goal (e.g. unloading a shipment), the system determines a sequenced set of actions (use the forklift to unload, cut the banderole, open the packages, etc.). This progression enables robots to evolve from executing isolated motions to performing coherent, multistep tasks, approaching human-level task intuition and planning capabilities. In essence, robots are enabled to “think” and plan tasks with a level of flexibility and context-awareness previously unattainable. Dexterous manipulation and mobility Advances in materials, actuators and robotic designs have greatly expanded what robots can physically do. Hardware breakthroughs – from high-precision force-controlled motors to soft robotic grippers – give machines much more dexterity in handling objects. Robots can now grasp irregular or delicate items reliably, rather than being limited to rigid, predefined motions. This is complemented by AI-driven control software that adjusts grip and force in real time. Notably, the incorporation of a sense of touch through modern tactile sensors is a primary enabler of human-level dexterity, allowing robots to finely manipulate objects through feedback of pressure and slip. Longer battery life is significantly increasing the uptime of mobile robots, supporting more autonomous deployments and leading to extended mobility. Moreover, robotics is no longer confined to traditional form factors. Innovations have introduced quadrupeds, humanoids, mobile manipulators and hybrid forms, broadening the range of industrial applications and increasing the scope of feasible automation. These physical innovations enable robots to “act” on the world with far greater skill and autonomy. 1.2 Enhanced capabilities enabling end-to-end automation These enhanced capabilities led to the evolution of robotics from (1) rule-based robotics that are explicitly programmed to (2) training-based robotics that acquire their skill in the real world and through simulation training to (3) context-based robotics performing tasks autonomously without explicit training through zero-shot learning. Advances in all three robotic systems transform operations and expand the automation scope to tasks that previously could not be automatable.8 At the heart of this transformation, however, lies the coexistence of all three foundational robotics systems, each expanding in automation scope and sophistication. Together, they form a complementary ecosystem. Rather than replacing one another, they enable a layered automation strategy, aligned with operational needs (e.g. degrees of task variability) and economic considerations. Furthermore, as factories and warehouses move towards greater automation, manufacturers and warehouse operators will deploy a mix of robotic systems and embodiments – from autonomous mobile robots (AMRs) to humanoids – guided by task requirements, economic viability and process characteristics. Physical AI: Powering the New Age of Industrial Operations 7

Physical AI Powering the New Age of Industrial Operations 2025