Humanoid robotics has moved from research labs to commercial roadmaps, with companies including Tesla, Figure, and Agility Robotics publicly targeting production within the next few years. For design engineers, that shift raises questions that don’t map neatly onto existing robotics experience.
Electronic Specifier recently brought together three engineers working across the humanoid supply chain to discuss the key design considerations shaping early-stage decisions: Martin Schiestl, System Application Engineer at Infineon Technologies; Kristen Mogensen, System Engineer for Industrial Systems – Robotics at Texas Instruments; and Aly Barakat, Technical Marketing Engineer at Mouser Electronics. Hosted by Managing Editor Paige Hookway, the webinar covered timelines, mechanical and thermal trade-offs, autonomy versus teleoperation, and functional safety.
Are the production timelines realistic?
Asked how achievable near-term targets are, all three gave a qualified yes – with the caveat that ‘production’ needs defining. For narrow, well-bounded tasks such as tote handling, Schiestl said two-to-three-year timelines are “reasonable and valid,” though progress on general-purpose humanoids is far harder to predict.
Barakat drew a sharper line between manufacturing capacity and genuine utility: “There’s a big difference between being able to manufacture a humanoid, deploying it on our customers’ side, and having it create reliable economic value day by day.” The harder problem, he argued, is getting a robot to work full shifts and recover from the unexpected without heavy human support: “I personally see the race won’t simply be who built the most robots, but how many productive autonomous hours do these robots actually deliver.”
Where will humanoids actually earn their keep?
There was strong consensus that the earliest meaningful deployments won’t be the flashy demonstrations that dominate social media. Barakat argued the real opportunity lies in material handling and intra-factory logistics – the gaps between existing “automated islands” designed around human workers. “Instead of redesigning the whole factory around a robot, you potentially build the humanoid that can operate inside the infrastructure that already exists,” he said. Schiestl agreed classical, special-purpose robotics “is here to stay” for high-volume tasks, with humanoids finding their niche where flexibility is needed – often in medium-sized companies rather than the giant fleets marketing suggests.
Mechanical and thermal trade-offs
The panel returned to a simple principle: mass is far costlier the further it sits from the robot’s centre. “Every gram which is further away from the central mass body … adds more to the torque,” Schiestl explained, citing higher switching frequency and smarter control as ways to claw back weight at distal joints. Barakat put it in blunt commercial terms: “Customers in general, they don’t want to buy joints. They just want to buy something that’s useful,” arguing designers should resist chasing human-like dexterity for appearance’s sake, since every added degree of freedom is its own failure point.
Gallium nitride (GaN) power stages featured heavily, with Mogensen noting TI has investigated GaN FETs since 2017 to boost power density. Schiestl, who won the Young Engineer Award at PCIM 2025 for a compact GaN-based motor drive, said the benefit flows through to reduced motor losses at system level: “We always need to look at the whole system to get the most benefit.”
Thermal management proved just as thorny. Where an industrial robot relies on external cabinets and heat sinks, a humanoid must carry it all internally. Mogensen flagged a hard constraint: surfaces above 50°C risk burning anyone who touches the robot, capping how much heat can be spread across the chassis. Schiestl suggested heat pipes and distributing thermal load into limbs, while Barakat noted compute generates continuous heat while joints spike during lifting or balance recovery – so cooling and sensing “have to be part of the design from the beginning.”
Autonomy, teleoperation, and the safety question
On the split between onboard autonomy and teleoperation, Barakat was clear that anything touching balance or safety must run locally: “We cannot depend on Cloud communication for stabilising a humanoid.” He sees teleoperation as a bridge to autonomy, but warned against a model where one operator babysits one robot: “If every robot needs one remote operator sitting around controlling it all day, then we didn’t really automate the job.” Mogensen agreed it isn’t viable for safety-critical response: “there has to be a local safety.”
Safety dominated much of the discussion. Mogensen highlighted a challenge unique to legged humanoids: unlike a bolted-down cobot with a holding brake, “if you have a humanoid on legs, if you remove power, it’s going to collapse. How do you do safe collapsing?” He pointed to draft ISO Working Group 12 standards on motor-drive safety, alongside a separate emerging AI standard, as critical for the industry to track. Schiestl agreed standardisation “gives us certainty,” predicting wheeled robots will achieve compliance before legged humanoids, since certifying every joint as fail-safe is complex. Barakat added that safety needs designing in from the start, “not later during the build of the robot.”
Advice for teams starting out
Asked what to nail down first, Barakat argued for defining the job and duty cycle before the robot itself: “What useful work does this machine actually need to deliver every hour?” Schiestl’s advice was to resist premature optimisation – standardising around three or four power levels across dozens of joints to ease the transition to production. Mogensen’s concern was sequencing when to add safety features, cautioning that “safety sometimes is very unforgiving on a hardware perspective.”
Looking ahead
Closing out audience questions, Barakat predicted humanoids in five years will be “more refinement than radical redesign,” with lighter actuators and fewer unnecessary joints. Schiestl suggested some industrial humanoids may not need heads at all, instead using interchangeable tooling like a worker swapping hand tools. Mogensen expects early iterations to last six months to two years as companies learn and iterate, improving once designs mature.
The panel’s closing message was consistent: engineering progress is real, but so is the gap between demo and deployment. As Mogensen put it, the path forward runs through “building the right safety story” to get real-world data – without which none of the harder autonomy problems can be solved either.
Watch the webinar on-demand below:
