We've trained an AI system to solve the Rubik's Cube with a human-like robot hand. This is an unprecedented level of dexterity for a robot, and is hard even for humans to do. The system trains in an imperfect simulation and quickly adapts to reality:
#DeepLearning #hype alert: ⁦⁩’s paper on Rubik’s cube solving starts “We’ve trained a pair of neural networks to solve the Rubik’s Cube with a human-like robot hand”; mentions critical hardwired symbol-manipulating cube solver only at ref. 111
OpenAI progress with physical robots over the past 2.5 years: - - - - - Pretty exciting to see the progress since our first result of a Spam Detecting robot:
We just released our scientific analysis of OpenAI Five: We are already using findings from Five in other systems at OpenAI like Dactyl () or our multi-agent work (). Hope that others find the results useful!
Amazing video "ADR frees us from having an accurate model of the real world & enables the transfer of neural networks learned in simulation to be applied to the real world" TY
Robots reach a major milestone: complex tasks such as solving Rubik's cube with one hand *learned* rather than programmed
Blogpost: "Since May 2017, we’ve been trying to train a human-like robotic hand to solve the Rubik’s Cube...we’ve reached our initial goal." Paper: The hand solves a cube with one sticker modified to reduce symmetry 20 percent of the time.
If AI can train a robot hand to solve Rubik’s cube, what scholcom problems can it solve? Find out at Thursday’s #FBF2019 Panel on AI in Academic Publishing #UNSILO
Cool work based on et al's POET (successful 60% of the time, 20% if majorly scrambled) #RL #simulation #metalearning Video:
Also, the videos and other materials in the blog post are gorgeous—OpenAI certainly knows how to make their work look pretty. I'd highly recommend clicking through and exploring.