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. openai.com/blog/solving-rubi…
1
1
9
Replying to @tsimonite
Getting beyond 0% success was the goal. Like with OpenAI Five, there's no fundamental barrier to increasing reliability (and note that after OpenAI Five's losses at TI 2018, many people thought that we'd hit the limits of our technology!).
Greg Brockman
 @gdb
15 Oct 2019
Our robotic system solves the cube 60% of the time under normal conditions — but only 20% with an adversarially-scrambled cube. The big result is that it's possible at all. Like with OpenAI Five, reliability keeps getting better the more we train

Oct 15, 2019 · 7:22 PM UTC

1
1
6
Replying to @gdb
Hi Greg! Yes, this is a tricky challenge. Your results are neat work. "Getting beyond 0% success was the goal." Are you going to keep working on Rubik's Cube or move onto other physical challenges?
1
Rubik’s certainly won’t be our next milestone, but I wouldn’t be surprised if it remained a benchmark!
2