AI Can't Outrun Humanoid Hardware Limits
RoboBusiness 2025 is set for Oct. 20-21 in Santa Clara. Andreas Friedrich, Managing Director of Technology & Strategy at Allegro MicroSystems, will discuss key technological breakthroughs enabling next-gen robotic actuation. RoboBusiness is organized by The Robot Report and its parent firm Arrowfly.
Humanoid robots become harder to build as their capabilities grow. Each articulating joint adds motors, sensors, and controls, all crammed into tight spaces while adding weight that can hinder motion. AI improves perception and decision-making, but translating those decisions into precise, safe movement is a distinct engineering problem. This gap widens when humanoids move from labs to real-world work environments.
Power architectures in advanced robots are shifting from low-voltage to 48V DC systems. Higher voltage cuts current fourfold, enabling lighter wiring and reducing resistive losses by 16 times. This slashes energy needs and simplifies thermal management in joint-packed machines.
However, higher voltage poses new risks. During fast deceleration, motors act as generators, producing voltage spikes far above the nominal bus level. Motor-control electronics must withstand such transients without failing or losing control. Every gain in power, weight, or heat brings tradeoffs affecting reliability and performance.
A high-level AI may tell a robot to pick up a cup or turn a handle, but each joint must still execute. Joints use closed-loop control, combining position, current, and torque sensors with continuous motor adjustments. Handing someone tea requires dynamic torque control and response to changing forces. This physical intelligence bridges AI decisions and controlled movement.
Dexterous hands and shoulders have different needs. Large motors create electromagnetic interference; small joints demand precise feedback in minuscule spaces. No single sensing solution fits all, and robust feedback is essential for converting machine intelligence into action.
When robots work alongside people, safety moves inside the control loop. Functional safety systems must detect faults and trigger safe states in milliseconds. Designers learn from automotive steering and braking, which embed diagnostics and safe-state behavior from the start. Integrated electromechanical subsystems—with sensing, control, power, thermal, and safety—become critical. Semiconductor integration shortens connections, reducing routing complexity and parasitic effects, letting engineers focus on robot-level software and behavior.
Touch and force sensing remain major challenges. Humans adjust grip based on feel, not predetermined trajectories. Robots need similar feedback to handle fragile objects. AI advances bring robots closer to deciding what to do; the next breakthrough depends on whether their physical systems can execute efficiently, precisely, and safely.