{"slug": "ai-cant-outrun-a-humanoids-hardware", "title": "AI can’t outrun a humanoid’s hardware", "summary": "Andreas Friedrich, Managing Director Technology & Strategy at Allegro MicroSystems, will speak at RoboBusiness on Oct. 20-21 in Santa Clara about the hardware challenges limiting humanoid robots, emphasizing that AI advances cannot overcome physical constraints. He highlights the shift to 48V DC power architectures to reduce wiring weight and losses, crucial for robots with 30 or more degrees of freedom.", "body_md": "**Editor’s Note:** Andreas Friedrich, Managing Director Technology & Strategy, Allegro MicroSystems, will be speaking at RoboBusiness (Oct. 20-21 in Santa Clara). His talk will dissect the critical technological advancements that are making next-generation robotic actuation possible. [Register for RoboBusiness today](https://www.robobusiness.com/). RoboBusiness is produced by The Robot Report and parent company Arrowfly.\n\nThe more capable a [humanoid robot](https://www.therobotreport.com/category/robots-platforms/humanoids/) becomes, the more difficult it is to accommodate the hardware required to make it move. The humanoid environment is one of articulated joints and degrees of freedom, where hardware faces stringent size and weight constraints, and motion requires a slew of motors, sensing, control, power and safety systems.\n\n[Artificial intelligence](https://www.therobotreport.com/category/design-development/ai-cognition/) has helped humanoids better perceive their surroundings, interpret instructions and determine what actions to take. Translating those decisions into precise, fluid and safe physical movement presents a different engineering challenge.\n\nThat distinction matters more as humanoids move from laboratory prototypes to practical machines capable of operating alongside people. Continued advances in AI matter, but the physical limits of the robot determine how that intelligence is put to work.\n\n## From fixed automation to autonomous machines\n\nConsider the difference between a traditional industrial robot and a humanoid. An industrial robotic arm may have six or seven axes and spend its entire working life bolted to a factory floor. Weight at its base may matter relatively little. Thick copper cables can deliver power. Motor drives can be housed outside the robot in control cabinets. If necessary, the system can accommodate substantial cooling infrastructure.\n\nA humanoid with 30 or more degrees of freedom has none of those luxuries. Every additional joint introduces another combination of motors, power electronics, position and current sensing, control and safety functions. All of that hardware has to travel with the robot. Every gram contributes to the energy required to move it, and every millimeter occupied by electronics competes for space inside an arm, leg, hand or finger.\n\nThe electronics required to give a humanoid freedom of movement take up precious space and add weight that can actually restrict how freely the robot moves. This engineering paradox is particularly acute in dexterous hands, where multiple motors, sensors and control functions must fit into extremely tight spaces, making the joint itself a system-level design problem.\n\n## When power architecture determines mobility\n\nThe evolving power architecture of advanced robots demonstrates how closely electrical and mechanical design are connected. Early robotic systems were commonly designed around relatively low-voltage power distribution. More advanced mobile robots and humanoids are moving toward higher voltages, specifically 48V DC architectures.\n\nPhysics explains why. For the same amount of delivered power, moving from 12V to 48V reduces the required current by a factor of four. Designers can use thinner wiring, reducing copper weight throughout the machine. And because resistive losses scale with the square of current, reducing current by four reduces wiring losses by a factor of 16.\n\nIn a humanoid, those electrical gains have mechanical consequences. Less wiring weight reduces the energy required for movement, while lower thermal loads make it easier to package electronics into joints that cannot accommodate elaborate cooling systems.\n\nHigher voltage, however, introduces its own engineering problem. Motors do not always consume energy. During rapid deceleration they can behave as generators, producing transient voltages considerably higher than the nominal DC bus voltage. The electronics controlling the joint must withstand those events without damage or loss of control.\n\nHigher voltage reduces current, heat and wiring weight, but it also raises the demands placed on motor-control electronics. Changes in power, thermal performance, weight or reliability can directly affect the others. Similar tradeoffs appear throughout the humanoid.\n\n## AI can decide, but the joint still has to execute\n\nA higher-level AI system may determine that a robot should pick up an object, turn a handle or hand someone a glass. But the instruction itself does not produce useful physical movement. Each joint must continuously determine where it is, how it is moving and how much force it is applying. Position, current and torque information feed a closed control loop that measures actual movement against commanded movement and continuously adjusts motor behavior.\n\nConsider the seemingly simple task of handing someone a cup of tea. The robot cannot just move its arm toward a predetermined coordinate. It must know the position of its joints and control the torque being applied as the arm moves, while responding to changing physical conditions. This form of physical intelligence encompasses the sensing, actuation and feedback required to translate an AI decision into controlled movement.\n\nMore sophisticated motion also places greater demands on sensing. A shoulder and a finger, for example, have very different mechanical and packaging requirements. Large motors can create electromagnetic interference that sensing systems must tolerate, while smaller joints impose severe space limitations and still require precise position data.\n\nNo single sensing architecture is likely to address every joint. Precise, reliable feedback is part of the control loop that turns machine intelligence into physical action.\n\n## Safety moves inside the control loop\n\nThat control becomes even more consequential when robots begin working alongside people. Industrial automation environments often manage risk by separating humans and machines. A humanoid designed to collaborate with a factory worker or assist with tasks in the home cannot depend on that separation. Functional safety becomes part of the robot.\n\nThe control system must continuously monitor operating conditions, detect faults or anomalous behavior and quickly move the affected subsystem to an appropriate safe state when necessary. Building those capabilities from scratch can create hardware redundancy, software overhead and certification work.\n\nRobotics designers can draw on engineering disciplines that have dealt with similar challenges. Automotive systems such as electronic steering and braking combine high-power electromechanical actuation with sensing, diagnostics and functional safety. Humanoids have different requirements, but the architectural lesson still applies: diagnostics, fault detection and safe-state behavior are easier to accommodate when designed into the system from the beginning rather than layered onto a finished design.\n\n## The joint becomes a system\n\nAt the scale required by advanced humanoids, sensing, motor control, power management, thermal performance and functional safety must be addressed as an integrated electromechanical subsystem capable of converting higher-level machine intelligence into controlled physical action.\n\nSemiconductor integration can shorten the physical and electrical distances between sensing, control and actuation, reducing PCB routing complexity, thermal hotspots and parasitic effects. Robotics engineers can then dedicate more time to the software, kinematics and system behavior that differentiate the robot rather than solving lower-level electronics problems within every joint.\n\nThe joint becomes a critical interface between AI and the physical world.\n\n## From seeing and moving to feeling\n\nToday’s humanoids use sophisticated vision systems and powerful AI models to recognize objects. But when a person grips a tool or handles a fragile object, touch and force feedback modify the movement as it occurs. We do not simply calculate the position of our fingers and execute a predetermined trajectory. We feel the object and adjust.\n\nReplicating human-like dexterity remains a significant challenge for robotics, particularly in the hand. Integrating force and touch sensing into fingers, fingertips and other surfaces gives machines information about what happens when they make contact with the physical world and allows them to adjust their movement in response.\n\nThe remarkable progress in AI has brought robotics closer to machines capable of determining what they should do. The next major advance may depend on whether their physical systems can execute those decisions efficiently, precisely and safely.", "url": "https://wpnews.pro/news/ai-cant-outrun-a-humanoids-hardware", "canonical_source": "https://www.therobotreport.com/ai-cant-outrun-a-humanoids-hardware/", "published_at": "2026-09-08 19:03:09+00:00", "updated_at": "2026-09-08 19:18:33.467061+00:00", "lang": "en", "topics": ["robotics", "ai-infrastructure"], "entities": ["Andreas Friedrich", "Allegro MicroSystems", "RoboBusiness", "The Robot Report", "Arrowfly"], "alternates": {"html": "https://wpnews.pro/news/ai-cant-outrun-a-humanoids-hardware", "markdown": "https://wpnews.pro/news/ai-cant-outrun-a-humanoids-hardware.md", "text": "https://wpnews.pro/news/ai-cant-outrun-a-humanoids-hardware.txt", "jsonld": "https://wpnews.pro/news/ai-cant-outrun-a-humanoids-hardware.jsonld"}}