AI Learns to Escape (deep reinforcement learning)
AI Learns to Escape (deep reinforcement learning)
AI Teaches Itself How to Escape!
In this video an AI named Albert learns how to escape 5 rooms I’ve designed. The AI was trained using Deep Reinforcement Learning, a method of Machine Learning which involves rewarding the agent for doing something correctly, and punishing it for doing anything incorrectly. Albert’s actions are controlled by a Neural Network that’s updated after each attempt in order to try to give Albert more rewards and less punishments over time.
Everything in this video (except for the music) was created entirely by myself using Unity. Check the pinned comment for more information on how the AI was trained!
Current Subscribers: 0
Content
0 -> This is Albert.
3.26 -> He's an artificial intelligence that learns.
5.804 -> He can move,
7.397 -> turn and jump,
9.668 -> and he has 10 seconds to escape this room
13.304 -> His movements start off 'random'
19.235 -> But when he does something good...
21 -> he is rewarded
23.124 -> and when he messes up
25.062 -> he is punished.
26.947 -> This is Room 1 of 5.
30.571 -> Albert, let's see if you can escape them.
35.235 -> You figured out how to open the door
40.071 -> but you can't do much else...
45.601 -> What about now?
51.413 -> Yes!
52.872 -> Now it's time for Room 2.
55.746 -> There are 2 pressure plates
57.548 -> So, Albert, you need to jump over the wall.
62.243 -> That's not how you do that...
70.019 -> Still not right...
79.467 -> Perfect!
80.697 -> That was too easy though.
82.708 -> Let's see if you can escape Room 3.
88.272 -> This one is much harder.
90.272 -> You need to learn to differentiate between
92.729 -> platforms you jump on, and walls you jump over
98.754 -> ...
100.442 -> not even close...
104.559 -> You hit the pressure plate, good job!
111.082 -> Again! But getting off is going to be tricky...
120.323 -> Walking off won't work, you need to jump.
125.251 -> Good!
127.136 -> Now you just need to find the door...
134.769 -> it's not over here...
143.663 -> Here we go
147.989 -> Why did you jump?!?
161.884 -> There we go!
163.395 -> There we go! (sorta)
164.239 -> This is Room 4
167.738 -> Now you need to learn to jump to different platforms
170.55 -> I'll give you 15 seconds to do this one.
175.371 -> It's going to be really difficult.
181.772 -> The tall platform will be especially tricky to get
185.574 -> come on, Albert, you got this
192.75 -> Nice!!
194.479 -> but it took way too long to get there...
198.772 -> You did it again!
203.395 -> Uhh...
205.208 -> there's still more you need to do...
208.363 -> ...
211.001 -> Let's try that one again, Albert.
222.209 -> Nice!
225.973 -> but you ran out of time...
231.324 -> Really?
248.849 -> You got them all!
250.266 -> But just barely ran out of time...
262.045 -> Finally!
265.211 -> This is the final challenge, Room 5.
268.096 -> You need to jump around platforms
270.762 -> to hit 6 pressure plates
273.553 -> then get down from the highest one
276.313 -> Good luck!
278.625 -> Room 3 taught you to jump on the first platform
281.519 -> but I think the wall's confusing you...
295.603 -> Why don't you try that again.
304.786 -> Okay, the walls are really messing you up.
311.773 -> You learned to jump away from the wall, that's a start
314.939 -> but you jumped into this dead end...
325.44 -> There's 2 pressure plates!
340.819 -> 3! (sorta)
343.631 -> What about now?
351.912 -> You're starting to get it!!
368.489 -> No, Albert, that's the wrong way...
380.742 -> You need to get the other one first...
385.751 -> for now though, 4 pressure plates isn't bad
388.4 -> but you're trapped...
402.001 -> Yes! You're back on track!
404.21 -> no...
411.401 -> You've had hundreds of thousands of attempts now
414.161 -> please, do something good.
424.08 -> Here we go!
426.788 -> nevermind...
450.746 -> So close!
479.064 -> Yes!
480.366 -> Congratulations, Albert!
483.22 -> But you didn't actually think you'd be able to escape, right?
487.084 -> No, I have a lot more planned for you :)
Source: https://www.youtube.com/watch?v=v3UBlEJDXR0