Learning to Walk in Minutes Using Massively Parallel Deep RL

İmpressive, very nice work

I think we need ant robots, the can carry hundreds times they weight...

This gave me anxiety

If I don't have to pick up their 💩💩 or exercise them, I'll take one!

what are the sensor to learn to walk? lidar? camera?

I am confused.
We already saw your quadropods like ANYmal being tested under real world conditions.
What is so different now, that you are back to virtual and climbing over cardboard boxes?

Why robots has legs? There is not more effective to use wheels?

What environment you use for simulation ?

Using massive clone jutsu learn things faster copied from baritone😂

boston dynamics be like "we just wasted 30 years of our lives"

What parameters make a robot in the iteration successful? Distance travelled?

Incredible Work!

Massive Parallel Deep RL aka Random Search

I get anxiety when I think about how a proper contact simulation for a single robot struggles to run in real time on my computer..

awesome! now do it with interacting robots (batching will no longer work?) :)

Geez, how many cpu cores does it take to run this many instances in parallel?

great