"In a new essay, he calls this approach "pure fantasy thinking." The problem? Human hands are incredibly sophisticated, packed with about 17,000 specialized touch receptors that no robot comes close to matching. While machine learning transformed speech recognition and image processing, those breakthroughs built on decades of existing technology for capturing the right data. "We don't have such a tradition for touch data," Brooks points out."
"Then there's safety. Full-sized walking humanoid robots pump massive amounts of energy into staying upright. When they fall, they're dangerous. Physics means a robot twice the size of today's models would pack eight times the harmful energy. Brooks predicts that in 15 years, successful "humanoid" robots will actually have wheels, multiple arms, and specialized sensors and abandon the human form. Meanwhile, he's thoroughly convinced that today's billions are funding expensive training experiments that will never scale to mass production."
Many humanoid robot startups train dexterity by showing robots videos of humans performing tasks, relying on visual imitation rather than rich tactile inputs. Human hands contain roughly 17,000 specialized touch receptors, a level of sensing that current robots do not approach. Machine learning advances in speech and vision built on long traditions of data capture that do not exist for touch. Full-size walking humanoids store large amounts of energy to remain upright and present severe safety risks when they fall; scaling up size dramatically increases harmful energy. Future practical designs are more likely to use wheels, multiple arms, and specialized sensors instead of a strict human form, while current large investments fund costly training experiments with limited mass-production prospects.
Read at TechCrunch
Unable to calculate read time
Collection
[
|
...
]