CASE STUDY 2 Electronics manufacturing (continued) Mission As part of Foxconn’s continuous drive for operational excellence and sustainable innovation, they found that traditional rule-based robotics fell short in automating precision tasks like screw tightening and cable insertion due to the need for high accuracy, adaptability and precise force control. Foxconn is innovating these precision tasks through robotics powered by AI and digital twin technologies. By integrating intelligent automation, real-time simulation and precision control, it enabled faster deployment, higher reliability and scalable implementation across global manufacturing operations – laying the groundwork for the next generation of smart, AI-integrated factories. Innovation in actionFoxconn used Nvidia’s platform alongside AI-powered robotic arms integrating precise socket pose estimation and real-time motion planning. This enabled highly accurate and collision-free operations, unlocking two new opportunities: –Screw tightening: AI-enabled robots learned optimal motion trajectories and torque application through reinforcement learning – improving consistency and cycle time, and reducing defects. –Cable insertion: Previously this could not be automated due to complexity, but it is now enabled through real-time force and trajectory adjustments – allowing dynamic grip and movement to accommodate part variation. The virtualization of the training and the integration of physical AI enabled fast deployment across multiple sites, forming a scalable blueprint for future Foxconn smart factories. Impact By creating digital twins of their production lines for rapid virtual simulation, testing and validation, Foxconn cut deployment time by 40%. AI-driven robotics improved cycle times by 20–30% and enhanced force feedback and motion control, reducing error rates by 25%. Virtual validation eliminated costly trial-and- error in physical environments, reducing operational costs by 15%. Moreover, the self-adjusting force and trajectory of the AI-driven robotic arms boosted precision and reliability, achieving a higher success rate than human workers in complex assembly tasks.1 Screw-tightening workcells simulation Simulation of a flexible workcell approach using different types of robotic arms.High-precision tasks powered by AI and simulation 2 Cable insertion Using two different approaches: manual and with robotic arms. Three workcells for robotic arm screw-tightening Manual cable insertion Single workcell for dual-arm cooperative screw-tightening Cable insertion with robotic arms Foundation and learningsTechnology: Simulation-to-reality transfer enabled large-scale deployment by accelerating AI model adaptation and reducing physical trial and error. People: Engineers were upskilled via hands-on training in digital twin simulation, AI-based robot programming and toolchains from Nvidia’s graphics collaboration platform Omniverse. This gave them skills in virtual commissioning, adaptive control and data-driven optimization, transitioning them from engineers to AI-integrated automation architects. Partnerships: Foxconn collaborated with service providers, such as Nvidia, to provide simulation and AI-powered robotics infrastructure. It also partnered with the manufacturing ecosystem (e.g. Fanuc and Techman) to co-develop scalable automation strategies, which were instrumental during this transformation. Physical AI: Powering the New Age of Industrial Operations 15

Physical AI Powering the New Age of Industrial Operations 2025