A commodity robot, especially substantially open source, could compete well with all of these products.
There are several aspects which might be subsystems or components that become open or closed commodities. A great hand - for a reasonable price. Good motors + drivers. Joint mechanisms. And the compute + main sensors (camera, positioning, audio, networking) - which might be a mobile phone and/or more.
For typical humanoid robots, there probably won't be auto company like vertical integration.
What will be nice, at some point, are standard interconnect +
power + communication points & subsystems - another potential
open or closed subsystem.
Stephen
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/16109887.1224388.1745480336728%40mail.yahoo.com.
|
Stephen D.
Williams
Founder: VolksDroid, Blue Scholar Foundation |
Training an LLM with all of the text, video, and audio knowledge
on the Internet (or at least a substantial fraction) is a massive
compute problem. I think better algorithms will cut that a lot
soon, but it will remain large. Training a robot to have
spatial-physical control, good basic eye / hand coordination is a
far simpler problem. A reasonably good vision system is more, but
that is reusable broadly. And that robot's system can leverage
the general intelligence of the big models without retraining,
only tuning & layering. We need more compute, but for the
robot maybe only something like 8 times the power of a Mac Studio
M3Ultra 512GB, for half the power (200w rather than <400w). Or
very fast low latency streaming to a nearby computer + cloud,
which is fine for a range of things. A nearby brain in a box is a
reasonable approach for a while, including in nearby vehicles.
The robot hardware is making progress, but still lacking good motor optimization plus much better mechanical design. And everything complex still costs too much. The Tesollo DG-5F is $17k each:
https://www.robotshop.com/products/tesollo-delto-gripper-5finger
Looks like that has a direct-drive servo in every finger joint: There is no room for any other linkage except hidden power + bearings. An interesting approach, but not viable for broad use. A robot hand should have 30-100 lb grip strength. I just checked: I can squeeze 64lbs with each hand.
Stephen
--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/d3c76931-d89f-4794-8c55-05e6a2e909d7n%40googlegroups.com.