Wow!! The demonstration where it learned to recover after being knocked over made me so excited! I think its so impressive that it was able to recover! Really neat stuff!
Very impressive. I've been trying to set up stuff like this (AI), but in a real time 3d environment, so physics are a little less real but this is really nice.
Essentially it got rewarded based on its performance, which encouraged good behavior and discouraged bad behavior. In the end it was very well behaved because it applied all the behavior that gave it the reward.
Other way and it would break. One of my professors showed us a video like this, and when he did it the other way, the machine didnt know that he was doing it in a different direction, that machine was a whole lot different, though. This one might be able to do it.
Wow!! The demonstration where it learned to recover after being knocked over made me so excited! I think its so impressive that it was able to recover! Really neat stuff!
I'm with you completely. The printer seems really pleased with itself.
Wow! It looks like the wobbliness in the frame is being compensated for like learned input shaping! that's awesome.
I think a system with less mechanical backlash would have performed better and learned more quickly.
0:56 my entire nerves system when my crush smiling at me
Wall-E
A little addition of LQR control will keep the pendulum on top.
Very impressive. I've been trying to set up stuff like this (AI), but in a real time 3d environment, so physics are a little less real but this is really nice.
so...in essence this machine learned to swing and balance the arm on its own?
That machine sounds like it's breaking itself apart
Hello.
Is it possible to find out which parts have been used for the mechanical
Amazing! made me laugh from excitement :D
What is the reward function for the cart inverted pendulum system?
Excellent, would love to see it do this on a mobile platform.
How was the performance of each trial assessed?
can i have the learning model where I can buy it
perfect =D
Essentially it got rewarded based on its performance, which encouraged good behavior and discouraged bad behavior. In the end it was very well behaved because it applied all the behavior that gave it the reward.
Other way and it would break. One of my professors showed us a video like this, and when he did it the other way, the machine didnt know that he was doing it in a different direction, that machine was a whole lot different, though. This one might be able to do it.
Any reason you only knock the pole down anti-clockwise?
Yes, but how did you do it??
hahahaha bro you hace to study control enginering and computer science for 7 years
@@juanfelipemontoya5821 easy i already done it but dont need computer science, need engineering
看起来效果很好。
proof of evolution
thats fucking cool
me he quedado anonadado
This is 10 years old, but hollly snap buddy fix your belts.
use PID for that king of tasks
reinforcement learning
But you have to tune a PID controller manually while this is automatic.