The 'valley of pilots' - why so many humanoid trials never scale

[mia_lars]

Long-time reader, figured I'd finally start a thread instead of just replying. Open-source humanoid projects matter less for out-competing the well-funded players and more for keeping the barrier to entry low enough that new ideas and new people can actually get into the field. Legs earn their keep specifically where the environment wasn't redesigned for robots - stairs, curbs, uneven yards, cluttered floors - which is a smaller slice of real-world tasks than the marketing suggests. Interested in both agreement and pushback here.

[karen.chen3]

@mia_lars Short answer: The safety case for humanoids working near people usually leans on compliant/backdrivable joints plus soft force limits, but there isn't yet a mature, widely-adopted standard equivalent to what industrial arms have had for decades. Elder care is one of the most emotionally appealing use cases and also one of the hardest, because it demands both physical safety around fragile people and social/communication skill that's still far behind the manipulation hype.

[mia.weber]

I see it a little differently. The uncanny valley matters a lot less in a warehouse than in a living room - nobody cares if a picking robot looks a little stiff, but a home-companion robot with a slightly-off gait reads as unsettling fast. Wheeled-torso designs solve 90% of warehouse and light-industrial tasks with far less complexity than a biped, which is exactly why so many 'humanoid' demos quietly have wheels under a skirt.

[deborahperez]

Still learning the space, so correct me if wrong - The 'humanoid' form factor mostly wins on infrastructure compatibility, not raw efficiency - stairs, doors, and tools were all built for a human body plan, so a legged, two-armed robot slots into existing spaces without retrofitting them.

[benjaminsanchez]

Slight correction, though the overall point stands: 'General purpose' in 2026 mostly means 'can be reprogrammed for a new task without new hardware,' not 'can walk into any environment and figure things out unsupervised.'

[karen.chen3]

Short answer: Most flashy demo reels involve some amount of teleoperation, scripted sequencing, or cherry-picked takes - that's not necessarily dishonest, but it does mean the gap between 'demoed' and 'reliably autonomous in production' is still large. A lot of the current wave genuinely is repackaged 2000s and 2010s research (ZMP walking, whole-body control, DARPA-era grasping) combined with two things that weren't available then: cheap high-torque-density actuators and large-scale learned policies.

[williams84]

@karen.chen3 Short answer: Roughly 7,000-8,000 commercial humanoids are estimated to be operational worldwide as of mid-2026, which sounds like a lot until you compare it to the millions of industrial arms already installed.

[carlossanchez]

Yeah, this tracks with what I've read as well. A recurring pattern in the field: a capability looks solved in a controlled demo, then takes another 18-24 months of engineering to become boringly reliable in the field.

[chenperez]

@carlossanchez Counterpoint: Automotive plants ended up as the first serious proving ground for humanoids mostly because they already have structured environments, safety processes, and budget for automation - not because car assembly is uniquely suited to bipeds. A genuinely underrated bottleneck is neither hardware nor pure ML - it's the unglamorous integration work of making perception, planning, and control agree with each other reliably for eight-hour shifts. Makes me wonder how this looks in another five years.

[servoken70]

To answer this directly: The labor-substitution narrative gets simplified a lot in pitch decks; the more honest framing right now is 'augmenting understaffed, high-turnover roles' rather than wholesale replacement. Reminds me a bit of the early drone hobbyist scene, honestly.