5 ms·

The objections to neural nets as the solution to all the problems on his list are the same as the old objections to neural nets as the solution to computer visi

The objections to neural nets as the solution to all the problems on his list are the same as the old objections to neural nets as the solution to computer vision, speech recognition, translation, playing Go, etc. The objections will fall in the face of overwhelming evidence that neural nets simply work better than other approaches to these types of problems. For a long time software was the reason robots didn't work, but that time is ending. Hardware will be the bottleneck soon if it isn't already. No general purpose robot will ever be successful in the mass market using electric motors and gearboxes at each joint. We need simpler, cheaper, lighter, more robust, more reliable, backdrivable, force-sensing actuators.

tuatoru 6y ago
> The objections will fall in the face of overwhelming evidence that neural nets simply work better Perhaps? But progress is stalling[1]. We need theoretical foundations. [1] https://www.sciencemag.org/news/2020/05/eye-catching-advances-some-ai-fields-are-not-real https://www.sciencemag.org/news/2020/05/eye-catching-advance...
- nmca 6y ago
  Wrg to your reference: One wonders why the "modest tweaks" weren't used in the original paper - indeed because they weren't known or weren't possible. The way you should interpet these meta-analysis is "researchers are motivated to attribute improvements in performance to 'the interesting bit' and they will run baselines without applying similar tweaking effort to them" The progress in the field can in some sense be indicated by the huge improvement in the 2006 method due to the "tweaks" discoverd since.
iamgopal 6y ago
Any research/potential example of such actuators ?
- vsskanth 6y ago
  Probably electro active polymers
- macrolime 6y ago
  Force can also be transmitted with belts, cables etc, like tendons. Then you don't need necessarily need a motor in each joint.
- cpgxiii 6y ago
  I would say some of the most promising work right now is in quasi-direct-drive actuators, basically low-gear-ratio brushless motors. Standalone electric motors are basically linear in terms of electrical power->torque, and even with a gearbox, provided the gear ratio is low enough, you can assume near-linear behavior. Brushless motors tend to be easier to design around specific size and shape constraints, and often run slower than brushed. The "secret" to a lot of this actuator work is that outside of highly repetitive industrial tasks, actuators are rarely running at full power so you don't need to design them (and their cooling) for continuous max power operation. Somewhat like human muscles, you can pick smaller motors and/or lower gear ratios that cover the normal cases, and still be OK with rare overloads so long as the overload isn't long enough to cause overheating.

tuatoru 6y ago

> The objections will fall in the face of overwhelming evidence that neural nets simply work better Perhaps? But progress is stalling[1]. We need theoretical foundations. [1] https://www.sciencemag.org/news/2020/05/eye-catching-advances-some-ai-fields-are-not-real https://www.sciencemag.org/news/2020/05/eye-catching-advance...

nmca 6y ago
Wrg to your reference: One wonders why the "modest tweaks" weren't used in the original paper - indeed because they weren't known or weren't possible. The way you should interpet these meta-analysis is "researchers are motivated to attribute improvements in performance to 'the interesting bit' and they will run baselines without applying similar tweaking effort to them" The progress in the field can in some sense be indicated by the huge improvement in the 2006 method due to the "tweaks" discoverd since.

iamgopal 6y ago

Any research/potential example of such actuators ?

vsskanth 6y ago
Probably electro active polymers
macrolime 6y ago
Force can also be transmitted with belts, cables etc, like tendons. Then you don't need necessarily need a motor in each joint.
cpgxiii 6y ago
I would say some of the most promising work right now is in quasi-direct-drive actuators, basically low-gear-ratio brushless motors. Standalone electric motors are basically linear in terms of electrical power->torque, and even with a gearbox, provided the gear ratio is low enough, you can assume near-linear behavior. Brushless motors tend to be easier to design around specific size and shape constraints, and often run slower than brushed. The "secret" to a lot of this actuator work is that outside of highly repetitive industrial tasks, actuators are rarely running at full power so you don't need to design them (and their cooling) for continuous max power operation. Somewhat like human muscles, you can pick smaller motors and/or lower gear ratios that cover the normal cases, and still be OK with rare overloads so long as the overload isn't long enough to cause overheating.