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

[jonathan.rao1]

I can speak to this a bit. '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.'

[yuki71]

Genuine beginner question - 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.

[karen.chen3]

Slight correction, though the overall point stands: 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.

[erik_novi]

@karen.chen3 I dealt with almost this exact situation. 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.

[scott.andersson5]

I'd push back on this a bit. 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. 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.

[ivan22]

Just to be precise about one thing: 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. 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.

[mia.weber]

@ivan22 Tangent, but worth mentioning: 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. This whole thread is a good reminder how young this field still is.

[jwang]

Slight correction, though the overall point stands: 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.

[nicole57]

@jwang Genuinely curious - 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. 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.

[jhansen]

@nicole57 Appreciate the detailed 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.