4 ms·

You'd think there would be more startups in this space given the demand. I understand robotics is hard, but money makes it easier.

AlexandrB 2y ago
Robotics is capital intensive and doesn't scale the way a software startup does. It's akin to starting a new car company but for a product category that has not yet been successful in the market - very risky. Then combine this with inevitable competition from China, who has dominated recent robotics-adjacent consumer product categories: drones and robot vacuums. I don't blame investors/founders for being hesitant. Edit: Oh yeah, and then there's safety/liability. A robot capable of doing laundry is also capable of breaking your finger(s) or poking your eye out. Just imagine the TikTok "challenges" kids would come up with interacting with such a device.
Grimblewald 2y ago
I dont think you understand just quite how much harder a laundry bot is than simply stealing everyones work and training a model to learn an approximation of the source distribution. Where would you even get the training data for starters? Not like people tend to wear mocap gear to do their chores yanno? Not like theyve been doing it the last few centuries, cataloging and digitizing everything dilligently. Solving washing machine bot would get us a lot closer to AGI than I think llm's / music / art bots will in the forseeable future.
- potatoman22 2y ago
  You make some good points, and I think I agree with them. I know robotics tends to be harder [1], but I don't understand how much harder it is. Visual-Language-Action models [2] have been gaining traction recently, along with training in simulated environments. Fine-tuning VLA models for specific purposes may reduce the data requirement. Plus, it's pretty cheap to record someone washing the dishes. [1] https://en.wikipedia.org/wiki/Moravec's_paradox https://en.wikipedia.org/wiki/Moravec's_paradox [2] https://openvla.github.io/ https://openvla.github.io/

AlexandrB 2y ago

Robotics is capital intensive and doesn't scale the way a software startup does. It's akin to starting a new car company but for a product category that has not yet been successful in the market - very risky. Then combine this with inevitable competition from China, who has dominated recent robotics-adjacent consumer product categories: drones and robot vacuums. I don't blame investors/founders for being hesitant. Edit: Oh yeah, and then there's safety/liability. A robot capable of doing laundry is also capable of breaking your finger(s) or poking your eye out. Just imagine the TikTok "challenges" kids would come up with interacting with such a device.

Grimblewald 2y ago

I dont think you understand just quite how much harder a laundry bot is than simply stealing everyones work and training a model to learn an approximation of the source distribution. Where would you even get the training data for starters? Not like people tend to wear mocap gear to do their chores yanno? Not like theyve been doing it the last few centuries, cataloging and digitizing everything dilligently. Solving washing machine bot would get us a lot closer to AGI than I think llm's / music / art bots will in the forseeable future.

potatoman22 2y ago
You make some good points, and I think I agree with them. I know robotics tends to be harder [1], but I don't understand how much harder it is. Visual-Language-Action models [2] have been gaining traction recently, along with training in simulated environments. Fine-tuning VLA models for specific purposes may reduce the data requirement. Plus, it's pretty cheap to record someone washing the dishes. [1] https://en.wikipedia.org/wiki/Moravec's_paradox https://en.wikipedia.org/wiki/Moravec's_paradox [2] https://openvla.github.io/ https://openvla.github.io/