Coding Challenge #29: Smart Rockets in p5.js

Videos:

Related Coding Challenges:

Timestamps:
0:00 Introduction
3:10 Rocket object
7:07 Population object
10:48 DNA object
15:09 Add target
17:24 Evaluate function
19:35 Mating pool
23:26 Selection function
25:36 Crossover function
33:56 Is the genetic algorithm working?
35:50 Check if rocket has reached target
38:05 Mutation function
40:17 Add obstacle
43:34 Check if rocket hits canvas edges
46:55 Conclusion and possible improvements

Editing by Mathieu Blanchette
Animations by Jason Heglund
Music from Epidemic Sound

#geneticalgorithm #dna #evolutionaryalgorithm #p5js #javascript

[kevnar]

A smart rocket walks into a bar. Daniel resets the simulation with a whole new population and tries again.

[LoganLovell]

coding challenges in the past: i will give myself 10 minutes
Looks at run time: 48:16
Oh boy, this must have been a good one

[mouse]

I love how enthusiastic you are in these videos. You're a great teacher!

[Salva]

This video got me really interested in genetic algorithms, thanks.

[ScubzMcTalBowling]

I made an email signature for my boss using HTML. Changed the color and size of font, etc. I felt accomplished.

[batfan]

This was interesting. Understood about 80% of it.

[bsizzbs]

Hey Dan, thanks for taking the time to do it from scratch! I think it really helps learn the subject but also helps with some of the syntax and tricks you can do with p5js. Keep up the awesome work!

[noonecares]

oh god... my smart rockets developed sentience... now they are questioning why they were made on the endless task of seeking for the target... never ceasing, always trying to get more fit

[chrisjones]

I’ve lost a couple hours on the library error. I’m trying to learn, but if I could edit libraries or describe their cross functionality with various APIs I would not need a tutorial. Aside from the problem with the library and/or text editor/browser issue, the content and enthusiasm is top notch. I’m a 42 year old guy in the throws of a career change in which I’m already behind, and the train is accelerating! Fortunately, dedication counts for a great deal.

[morphman]

An interesting thing happened when I added time to the fitness value ((1/count) * 100 added to the fitness value before the 10 times multiplier whenever a rocket has reached the target).

After about 15 generations, the average fitness keeps bouncing between 1700 and 2100. The top fitness is 4010, but this is unattainable, as it would require the rocket to take less than a single frame to get from start to target. The top achievable right now (as far as I could tell) is 4006.7, with the record I've seen over a few hundred generations is 4006.66.

It's like when you add a second quality to fitness, it tries to incorporate that so hard that it will pick the faster rockets over the more accurate ones, until it notices that the faster ones miss the target, but those that hit the target are slower so they won't be picked as often. That drags the average down, as most rockets will miss the target, but those that hit will have a higher breeding chance as their score is more than 10 times those that miss (~380 compared to over 4000). So the more accurate ones are picked again, but the faster ones score slightly higher now, so they will start getting picked, reducing accuracy and so on and so forth in infinitum.

So past gen 15, I ended up every single time with about half the rockets hitting but slowly and half missing, but they shoot off quite quick. The mutation kicks in some times, creating a rocket that goes way off course, then that one is removed from the gene pool thanks to the evolution thing.

[CaptainAardvaark]

The ending has me getting all philosophical. Like what if one side were a bit shorter, but all the rockets went around the other side anyway, and kept doing that because it was what worked first, even though there was a better way.

[MrKraignos]

And this, kids, is what a real programmer is. Using his knowledge in science to simulate the existing world, then pushing further in order to solve problems. (Here it's Physics mixed with Genetics, could also be applied with knowledge and problems in Economics, Sociology, and so on...)
Subscribed.

[sayochikun]

if you calculate d with "1/(distance+1)", you will get a normalized exponential fitness range and eliminate divide by 0 posibility. which is great and easy

[flamexode]

Rocket flies through rocket
Ah yes the physics system is done

[samfrostinjapan]

I think a big reason they were all loving crashing was because the value was set to 1 for crashing, which about as good as if they got within a pixel of the target.

[JinTsen]

He forgot to revisit the other website to see how the rockets evolved ^^

[divadbate]

please please please make this project an ongoing series. it would be awesome to refine it over time and maybe end up with a population that can navigate a maze or something.

[heruka]

You genuinely seem like a good person, I love your enthusiasm. Keep it up!

[Timotheeee]

this was my first ever introduction to machine learning back in the day

[sumanpahari]

Daniel you are amazing And it helps my son out a lot. Thanks man.