AI Learning Morphology and Movement…at the Same Time!

29 thoughts on “AI Learning Morphology and Movement…at the Same Time!”

1. 2:25 omg those tiny lil legs going pitpatpitpat <3 great paper!

2. Where can i find its github or something similar

3. noooo you ruined the end of the video, where's the "i'll see you NEXT TIME"?

4. When you skip leg day

5. one of these days they'll be learning morphology, longevity, incept dates

6. I'm wondering. In real biology, powerful or large features require more resources (as in food energy) to grow and maintain.

   For example, humans never evolved the strength to break rocks or bend steel because the cost/benefit ratio of having such strength didn't allow for it. It would be a lot harder to take in enough food energy to maintain that kind of strength and it wasn't important enough for our survival. Sure we could have survived better with it, but we survive well enough without it to not have needed to go any further. We evolved our brains instead.

   Has anyone done an experiment like this where that mechanic was simulated? Having virtual "food" that the organism must consume to maintain energy levels, and larger or stronger body parts requiring more energy to maintain?

7. I'd like to apply for that position. My experience is watching 2 minute papers videos. See you in Vienna!

8. It looks like one of the Don Hertzfeldt's animations – Meaning of Life: https://www.youtube.com/watch?v=xMsyOowMaEY

9. Great work as usual, happy new year.

10. You walk with your brother.Suddenly,you encounter a troll who has captured an innocent person.
    He agree to let him go if you say something true.But before you can say anything,he capture your brother.And say" I will only free your brother if you say something false.You have only one affirmation to do,five word,and if what you're saying result in an unsolvable paradox who make me lie no matter what I do,I will eat everyone. "

    What do you say?

11. 3:15 IA does not give a damn, just wants your score at any cost hahahahah

12. As an arachnaphobe, the way the last ones move really bothered me. Super cool though

13. Great year, my friend! Rest well and many blessings!!!

14. What is the benefit? Cornell University's sea fish did the same in the real world some years ago.

15. And I see you next time!
    I always wait for that sentence in your videos…

16. So given all other options, the AI basically prefers to build a straight line to the goal. Seems smart to me.

17. mhh pretty good

18. 13th and 14th monthly salary. Yeah, that's definitely Austria. 😉
    I've done my master's with neural networks 10-15 years ago. There was a bit of a hype back then that was lacking a little in results. But with the increase of computing power in the last decade it's amazing to see how much has changed and how many real world applications are used nowadays.

19. I visited Vienna this summer and accidentally walked past TUWien! I didn't realize it was so centrally located. Made me think of you 🙂

20. Interesting that they're trying to teach an algorithm to complete a scenario only once, and then constrain its design to operate more the way they'd expect it to.

    In reality, we learn things to make them trivial in repetition. It makes no sense to cheat for a learning algorithm's sake unless the designer is incompetent.

    Forcing it to complete a minimum of 2 timed heats under the conditions that 1. the 2nd heat begins with the agent in the same orientation at its ((completion or time expiration) and physics rest) of the previous heat, 2. with the agent payload centered at the starting line, and 3. that the AI should have a rule with a bias towards creating agents that can successfully earn points in both heats, would cause every "falling giant" version to fail even without constraints. This would be a more accurate representation of a useful reality than a single sprint.

21. Great video as usual!

22. Related content that could interest people:
    – "Evolved virtual creatures" (1994): https://www.youtube.com/watch?v=JBgG_VSP7f8
    – "Evolving virtual creatures" (2007): https://www.youtube.com/watch?v=0_8tNGKm87U

23. @Karoly you have to review this: https://youtu.be/kSLJriaOumA – it's just blows my mind

24. Makes no practical sense without the energy consumption. Should have been one of the primary goals there. Still moves better than those creepy metal dogs for the military.

25. I'm using this technique to calibrate my trading bot. The price movement is like a terrain and the goal is to get to the end with as much profit as possible while minimizing the number of losing trades. I'm quite surprised to find it works so well, when re-calibrated often to suit the new market conditions. it re-combines about 8 parameters in hundreds of thousands of combinations to find what method would produce a good chance of making gains. But now I want to test how often it should calibrate and this means I need to run a simulation that runs a simulation . . .

26. Wish I had what was needed for this position. That would be cool for sure.

27. Na kedves Károly, kapaszkodj a székedbe, ha esetleg még nem láttad volna: https://arxiv.org/pdf/1812.02246v1.pdf 🙂

28. It would be awesome to add max energy consumption as a choice

29. That's really impressive.

Comments are closed.