Nostr Web Client

Humanoid robots are interesting, but over the next decade, keep in mind the power requirements.

The human brain is an *immensely* powerful computer and runs on like 20 watts. It is a breathtakingly efficient thing. Plus with a bit of water and (optionally) some food, a human body can operate for days.

Getting humanoid robots to the point of being able to walk around and participate in our world with high levels of processing and long run times is going to be a hard and long-term engineering challenge. Harder than EV adoption. Robots in the real world are an order of magnitude more complex to get right than robots in a controlled industrial setting.

I’m pretty bullish on smaller non-humanoid home robots though. Things like robodogs. They can plug themselves in to recharge whenever they are low on power, and can do all sorts of tasks that a well-trained dog could do, plus some other things (language recognition, a mounted arm that can grasp things better, etc). They’ll get exponentially better in the years ahead.

exist270 8mo ago 💬 1

Nah, the smart companies will employ some form of solar skin for trickle charging & kinetic energy return, which will give them an essentially limitless runtime in a FAR faster timeframe than anyone realizes. They'll self-regulate their power supply by "going outside for a break" or "staring out the window".

Essential tasks like obstacle avoidance, etc. can simply be package upgraded daily while charging; they will all be collectively learning in real-time & they will all experience incremental upgrades on a daily basis.

All the hard processing will be centralized to a data center to make it more efficient. The 'bots will simply be single nodes in a hive mind where all bots can upgrade one another via real-world data discovery. They will also utilize AI to train scenarios in a 3D environment at a rate that far exceeds real-world learning scenarios. The training models will have real-time feedback/input from every node & the hive mind will train THOUSANDS of possible solutions per second, exporting the most successful outcomes back to the hive mind in a never-ending recursive loop.

Nothing stops this train. 🤙

Reply to this note

Please Login to reply.

Discussion

Lyn Alden 8mo ago 💬 1

Efficient solar skin is likely more than a decade away. And the power it gets from just part of the surface of a human shaped body is little.

Calling back to the hive mind will indeed be a main way that robots solve complex things, but wireless data is imperfect and so there will be a significant percentage of issues where robots get stuck.

The more decades you look out, the more solvable those things are, but my post was about the next decade.

exist270 8mo ago

Fair game. 🤙