What is the primary blocker to achieving capable household robots today?

Aidan Gomez · Aug 19, 2024 · hedged

Robotics is where the big AI breakthroughs will come, and someone will crack cheap, robust general-purpose humanoid robotics within roughly five to ten years

Costs are already coming down and foundation/language models have produced planners that are far more dynamic and able to reason about the world, replacing brittle hard-coded task programming tied to exact environments

48:41 Foundation model planners will crack cheap general purpose humanoid robotics within five to ten years

Tom Hulme · May 8, 2024 · hedged

Robotics, which he long considered nearly uninvestable because applications were too narrow and specialized, may now be genuinely investable as robots become generalizable

Just as AI went from specialized to generalizable, the combination of computer vision, large language models, multimodality and cheaper componentry may make robots generalizable

70:42 Generalizable robots driven by ai advances make robotics investable now

Harry Stebbings · Apr 18, 2025

Fine-grained hand dexterity — knowing how much tension to apply to a glass, plumping a pillow — is a core unsolved blocker for household robots

30:44 Fine grained dexterity control is the core unsolved blocker

Richard Socher · Apr 18, 2025

Household robotics faces the same long-tail problem as radiology: single narrow capabilities are easy to superhuman, but covering the tail requires enormous data and resources

One radiology classifier can beat a human at one thing, but full end-to-end automation and FDA certification requires seeing the whole long tail of cases

30:58 Long tail data and resource requirements mirror radiologys automation problem

Harry Stebbings · Apr 18, 2025

LLMs and robotics have inverted bottlenecks: with LLMs the technology is ready and human adoption is the constraint, while with robotics demand exists but the technology is the constraint

People would readily adopt a robotic cleaner in every household if it worked

31:28 Technology not demand is the constraint unlike llms where adoption lags

Richard Socher · Apr 18, 2025 · hedged

Household robotics is now doable, but only through a lot of hard engineering and research work

Building blocks like Meta's open-sourced pressure-sensing fingertips exist and need to be incorporated

31:42 Household robotics is achievable with more engineering not fundamental breakthroughs