Collaborative Robotics CEO Brad Porter on why humanoids fail commercially and what mobile robots actually do at Maersk today

Summary

Brad Porter, founder and CEO of Collaborative Robotics, argues that the most commercially viable robotics opportunity right now is not humanoids but wheeled mobile robots doing unglamorous logistics work.

Porter recently returned from a live deployment with Maersk near Seattle, where Collaborative Robotics is helping with transload operations at the port. The work involves moving industrial carts loaded with up to 1,500 lbs of retail goods unloaded from ocean freight containers and staged for dispatch to big-box stores across the Pacific Northwest. The robots move the carts, relieving workers from hauling heavy loads.

Why humanoids fail

Porter studied humanoids deeply in 2018 while leading robotics at Amazon. His team identified 40 use cases where a humanoid could theoretically add value but concluded that none actually required a humanoid form factor. Wheeled robots move faster, are simpler, and are far more reliable.

The fundamental problem is physics. Generating strength from rotational motors in a human-scaled joint is inefficient. Humanoid manufacturers hand-wind motors and push current to the maximum, achieving only around 60% of human strength while burning out motors quickly. Boston Dynamics' pneumatic Atlas could do backflips because pneumatics deliver roughly 10x the power of electric motors, but pneumatic systems cannot be made reliable enough for production environments. Electric humanoids are caught between motors too small to be strong and motors too large to fit the form factor.

Porter also dismisses near-term consumer humanoid use cases. Emotional companion robots and novelty interactions with children are plausible niches, but meaningful domestic utility, anything requiring real grip strength, runs into the same motor problem.

Navigation and LiDAR costs

Autonomous navigation in commercial indoor environments is largely solved. Robots using LiDAR-based SLAM and stereo depth cameras can localize, detect humans, and navigate around obstacles at roughly human walking speeds. That capability works reliably in hospitals, airports, and warehouses today.

LiDAR costs have dropped sharply from roughly $50,000 per unit to $3,000–$6,000. That remains expensive for consumer devices but sits well within industrial robot economics when the robot generates meaningful labor savings. Porter expects costs to continue falling but says the current price is not a blocking issue for the deployments Collaborative Robotics targets.

The manipulation problem

Manipulation remains unsolved. Dexterity required to interact with the physical world at the level of opening an AirPod case is still beyond current systems. The industry uses simulation training, multi-robot data pooling, and teleoperation to build pre-trained base models, but Porter draws a sharp analogy to LLMs. Simulation and curated datasets establish statistical correlations, the equivalent of next-token prediction, but cannot cover the long tail of real-world edge cases.

Handling a door knob that turns the wrong way or sits two inches higher than expected requires real-world self-play, the same trial-and-error adaptation humans use instinctively. The field has not cracked that. Nearly all current research effort concentrates on pre-training. Bridging to genuine real-world generalization remains the open problem.