HUD, SimRunner, and Boid Energy

Added a HUD because I often found myself wanting to know what generation I was looking at or how many food sources had been consumed.

Adding SimRunner didn’t go so well so I returned it to the backlog for now. I’ll need to sort that out eventually because Simulation is gonna develop some weird coupling if I keep adding information about a generation’s run to it.

Stopped displaying the avatar unless the user moves it.

Okay, enough faffing about with small beans. Back to boid energy.

Huh. I can’t register MainVisualize as an entity listener because I don’t have a pointer to it. shared_from_this() doesn’t work. I get a runtime error saying it’s a bad weak pointer because nothing has a non-weak pointer to it. It’s just instantiated straight up in main.cpp.

SimRunner? Lol.

Lambdas?

SimRunner (Finally)

#

I went back to SimRunner. I just implemented wrappers for every Simulation method called except getters. resetAgents(), fastForward(), step() and also Evolution::selectAndMutate(). Then I moved the logic from Simulation and Evolution into SimRunner one method at a time, making sure everything worked in between. Eventually I figured it out.

I also had MainVisualize and MainEvaluate instantiate SimRunner using the same method as SimRenderer, using a static construction method that makes a shared pointer and returns it. That way SimRunner can also be registered as an IEntityListener.

After all of that I now have one class that contains all the information about setting up a run, stepping or fast-forwarding through it, and figuring out when the run is over. Booyah.

Boid Energy Implementation

#

The new setup with SimRunner allowed me to handle dying boids much more neatly. When boids die, I add their AgentProps to a list of deadAgents. For the generation fitness report I append the dead agents to the living ones but they go to the bottom because dying gives a -100 penalty. For selection I append dead agents until there are enough for selection.

After getting everything working it was time to experiment again. Wute!

Agents were usually using up all their energy and dying before they reached a single food source. I reduced the cost of movement significantly and they were still having trouble. They tend to start off going in circles so they still died before reaching food. Lots of wasted generations where they just explored the “no food zone” over and over until they died. So I shrunk the no food zone.

That helped a bunch, now one or maybe two boids would reach food and have the opportunity to continue exploring. But they usually wouldn’t get two and most of the boids were still dying without reaching any food, even after many generations.

Finally, I increased the amount of food. From 10 to 30 then to 60. That’s a lot of food but with these new energy boids it seems to work great. They often develop an affinity for food really early. And a somewhat slower speed, which makes sense. The energy cost of movement increases exponentially with speed.

Boid Energy Results

#

It’s cool to see how much of an impact energy has. Without it, going in circles with minor adjustments here and there (including turning away from food the agent happened to encounter) was a reasonable strategy. That was why I had to reduce the amount of food sources from 30 down to 10. But even then it took 30-40 generations before boids really developed the behaviour of turning toward food. Now it often happens in under 10.

(Those claims are based on recollection of runs… would be nice to have actual data. Like a script that performs a bunch of runs and reports on some aggregated metrics.)

Another nice side effect is that each generation only takes 500-700 steps, particularly at the beginning where boids are dying pretty quickly. So a run of 30 generations takes about 30 seconds. Sweet.

• L for Load is not clearing the deleted agents..?

Example of a Bad Run

#

Hmm, sometimes it’s still pretty ugly, though. It seems like these ones evolved not to get food but to just move as slow as possible so they wouldn’t get the penalty for dying. I guess they didn’t find enough food during their runs to figure out a better strategy. Oof.

Examples of Good Runs

#

And then you get a great run like this one:

Ooh, or even better look at this: This is how their weights look:

New Strategy? Narrowly Colliding

#

The weight in the last column steers toward food more as the value gets bigger. Interesting that these did not go all the way to one. I wonder if catching the food at the edge is actually better so that the boid has a chance of picking up another piece of food nearby which it wasn’t aware of. The boids are only given information on the nearest food, not all of the food within a particular range. The boids at the top of the screen in that last example are steering straight toward the food but there are a few in the bottom that seem to steer to pick up the food by hitting it with their sides instead.

And theoretically that’s the part of this that’s gonna be the most fun. Seeing the agents develop unexpected strategies.

Ideas for Later: Another Layer of Automation

#

Would be nice to do several runs of 30-100 generations and get a report showing each run’s performance, how they converged (or didn’t converge) on a certain level of performance, how often runs were good and how often they were horrible, etc.

Then, I’d like to tweak things like the amount of food, the amount of boids, the energy cost of movement, etc. and be able to compare them easily. Especially if I can compare several different values for one or two of these parameters. Sometimes the parameters might be dependent so that’d help a lot.

Like, here’s another bad one:

How often are they this bad? And what happened to make them so horrible? Anything I can change in the environment to encourage them to overcome their initial random weights?

I guess I could increase the population. With more boids, it’s more likely that at least one of them will have a reasonably good neural network…

But yeah. I’m increasing the value now to see what it looks like but I have no idea if the run I’m about to look at is an exception or

oh cool, in this environment it’s much more about speed. All the boids have adapted to move as quickly as possible rather than the 1/3 speed I saw in the runs with 10 boids.

Surprisingly, it only took 40 seconds to run 30 generations. I guess they were finishing 80% of the food very quickly so.

Well that’s much more fun, I think I’ll keep that. Haha.

I ran a couple more runs of 30 generations and got similar results.

100 Generations with 20 Energy Boids and 60 Food

#

Total generations: 100
Running generation 0... complete (1.61472 seconds, 660 steps, 2 consumed, 0.004 food/step)
Running generation 1... complete (1.70316 seconds, 721 steps, 4 consumed, 0.014 food/step)
Running generation 2... complete (1.47253 seconds, 717 steps, 5 consumed, 0.014 food/step)
Running generation 3... complete (1.50919 seconds, 741 steps, 4 consumed, 0.014 food/step)
Running generation 4... complete (2.90381 seconds, 1838 steps, 8 consumed, 0.004 food/step)
Running generation 5... complete (3.18798 seconds, 2535 steps, 15 consumed, 0.014 food/step)
Running generation 6... complete (2.29316 seconds, 974 steps, 2 consumed, 0.004 food/step)
Running generation 7... complete (3.00336 seconds, 1888 steps, 18 consumed, 0.014 food/step)
Running generation 8... complete (2.90838 seconds, 2828 steps, 48 consumed, 0.024 food/step)
Running generation 9... complete (2.38283 seconds, 1930 steps, 48 consumed, 0.024 food/step)
Running generation 10... complete (3.08518 seconds, 4275 steps, 42 consumed, 0.014 food/step)
Running generation 11... complete (2.29815 seconds, 1606 steps, 48 consumed, 0.034 food/step)
Running generation 12... complete (2.1177 seconds, 1384 steps, 48 consumed, 0.034 food/step)
Running generation 13... complete (1.95995 seconds, 1250 steps, 48 consumed, 0.044 food/step)
Running generation 14... complete (1.95804 seconds, 1397 steps, 48 consumed, 0.034 food/step)
Running generation 15... complete (2.34007 seconds, 2282 steps, 48 consumed, 0.024 food/step)
Running generation 16... complete (2.50734 seconds, 2675 steps, 48 consumed, 0.024 food/step)
Running generation 17... complete (2.45197 seconds, 2425 steps, 48 consumed, 0.024 food/step)
Running generation 18... complete (1.69949 seconds, 1160 steps, 48 consumed, 0.044 food/step)
Running generation 19... complete (1.96259 seconds, 1284 steps, 48 consumed, 0.044 food/step)
Running generation 20... complete (1.80562 seconds, 1196 steps, 48 consumed, 0.044 food/step)
Running generation 21... complete (2.35467 seconds, 2034 steps, 44 consumed, 0.024 food/step)
Running generation 22... complete (2.34795 seconds, 2272 steps, 48 consumed, 0.024 food/step)
Running generation 23... complete (1.79105 seconds, 1155 steps, 48 consumed, 0.044 food/step)
Running generation 24... complete (2.10706 seconds, 1476 steps, 48 consumed, 0.034 food/step)
Running generation 25... complete (2.33204 seconds, 2090 steps, 48 consumed, 0.024 food/step)
Running generation 26... complete (1.82486 seconds, 1260 steps, 48 consumed, 0.044 food/step)
Running generation 27... complete (1.64004 seconds, 969 steps, 48 consumed, 0.054 food/step)
Running generation 28... complete (1.73568 seconds, 1417 steps, 48 consumed, 0.034 food/step)
Running generation 29... complete (1.76338 seconds, 1200 steps, 48 consumed, 0.044 food/step)
Running generation 30... complete (1.6216 seconds, 972 steps, 48 consumed, 0.054 food/step)
Running generation 31... complete (1.8331 seconds, 1313 steps, 48 consumed, 0.044 food/step)
Running generation 32... complete (1.75284 seconds, 1311 steps, 48 consumed, 0.044 food/step)
Running generation 33... complete (1.76011 seconds, 1294 steps, 48 consumed, 0.044 food/step)
Running generation 34... complete (1.66547 seconds, 1322 steps, 48 consumed, 0.044 food/step)
Running generation 35... complete (1.64468 seconds, 1102 steps, 48 consumed, 0.044 food/step)
Running generation 36... complete (1.70166 seconds, 1159 steps, 48 consumed, 0.044 food/step)
Running generation 37... complete (1.73468 seconds, 1298 steps, 48 consumed, 0.044 food/step)
Running generation 38... complete (1.68521 seconds, 1155 steps, 48 consumed, 0.044 food/step)
Running generation 39... complete (1.51367 seconds, 971 steps, 48 consumed, 0.054 food/step)
Running generation 40... complete (1.68841 seconds, 1363 steps, 48 consumed, 0.044 food/step)
Running generation 41... complete (1.41528 seconds, 936 steps, 48 consumed, 0.054 food/step)
Running generation 42... complete (2.14678 seconds, 2051 steps, 49 consumed, 0.024 food/step)
Running generation 43... complete (1.79262 seconds, 1311 steps, 48 consumed, 0.044 food/step)
Running generation 44... complete (1.96798 seconds, 1987 steps, 48 consumed, 0.024 food/step)
Running generation 45... complete (1.70017 seconds, 1318 steps, 48 consumed, 0.044 food/step)
Running generation 46... complete (2.3044 seconds, 2660 steps, 47 consumed, 0.024 food/step)
Running generation 47... complete (1.8235 seconds, 1375 steps, 48 consumed, 0.034 food/step)
Running generation 48... complete (1.87262 seconds, 1817 steps, 48 consumed, 0.034 food/step)
Running generation 49... complete (1.55358 seconds, 1236 steps, 48 consumed, 0.044 food/step)
Running generation 50... complete (1.50649 seconds, 1080 steps, 48 consumed, 0.044 food/step)
Running generation 51... complete (1.70495 seconds, 1304 steps, 48 consumed, 0.044 food/step)
Running generation 52... complete (1.94269 seconds, 2436 steps, 41 consumed, 0.024 food/step)
Running generation 53... complete (1.77772 seconds, 1761 steps, 48 consumed, 0.034 food/step)
Running generation 54... complete (1.75251 seconds, 1852 steps, 48 consumed, 0.034 food/step)
Running generation 55... complete (1.59666 seconds, 1137 steps, 48 consumed, 0.044 food/step)
Running generation 56... complete (1.49455 seconds, 1140 steps, 48 consumed, 0.044 food/step)
Running generation 57... complete (1.31031 seconds, 810 steps, 48 consumed, 0.064 food/step)
Running generation 58... complete (1.57577 seconds, 1284 steps, 48 consumed, 0.044 food/step)
Running generation 59... complete (1.35141 seconds, 1031 steps, 48 consumed, 0.054 food/step)
Running generation 60... complete (1.5274 seconds, 1204 steps, 48 consumed, 0.044 food/step)
Running generation 61... complete (1.5673 seconds, 1320 steps, 48 consumed, 0.044 food/step)
Running generation 62... complete (1.56065 seconds, 1227 steps, 48 consumed, 0.044 food/step)
Running generation 63... complete (1.46666 seconds, 1024 steps, 48 consumed, 0.054 food/step)
Running generation 64... complete (1.34223 seconds, 972 steps, 48 consumed, 0.054 food/step)
Running generation 65... complete (1.46964 seconds, 1021 steps, 48 consumed, 0.054 food/step)
Running generation 66... complete (1.14699 seconds, 752 steps, 48 consumed, 0.064 food/step)
Running generation 67... complete (1.32553 seconds, 893 steps, 48 consumed, 0.054 food/step)
Running generation 68... complete (1.2531 seconds, 805 steps, 48 consumed, 0.064 food/step)
Running generation 69... complete (1.30491 seconds, 944 steps, 48 consumed, 0.054 food/step)
Running generation 70... complete (1.28459 seconds, 891 steps, 48 consumed, 0.054 food/step)
Running generation 71... complete (1.35237 seconds, 1056 steps, 48 consumed, 0.054 food/step)
Running generation 72... complete (1.20148 seconds, 748 steps, 48 consumed, 0.064 food/step)
Running generation 73... complete (1.14546 seconds, 790 steps, 48 consumed, 0.064 food/step)
Running generation 74... complete (1.19284 seconds, 987 steps, 48 consumed, 0.054 food/step)
Running generation 75... complete (1.48537 seconds, 1559 steps, 48 consumed, 0.034 food/step)
Running generation 76... complete (1.35841 seconds, 1427 steps, 48 consumed, 0.034 food/step)
Running generation 77... complete (1.3722 seconds, 1138 steps, 48 consumed, 0.044 food/step)
Running generation 78... complete (1.25716 seconds, 890 steps, 48 consumed, 0.054 food/step)
Running generation 79... complete (1.17233 seconds, 756 steps, 48 consumed, 0.064 food/step)
Running generation 80... complete (1.03586 seconds, 639 steps, 48 consumed, 0.084 food/step)
Running generation 81... complete (1.27294 seconds, 1028 steps, 48 consumed, 0.054 food/step)
Running generation 82... complete (1.09951 seconds, 784 steps, 48 consumed, 0.064 food/step)
Running generation 83... complete (1.22574 seconds, 887 steps, 48 consumed, 0.054 food/step)
Running generation 84... complete (1.46581 seconds, 1050 steps, 48 consumed, 0.054 food/step)
Running generation 85... complete (1.02203 seconds, 562 steps, 48 consumed, 0.094 food/step)
Running generation 86... complete (0.998088 seconds, 653 steps, 48 consumed, 0.074 food/step)
Running generation 87... complete (1.27174 seconds, 912 steps, 48 consumed, 0.054 food/step)
Running generation 88... complete (1.17371 seconds, 770 steps, 48 consumed, 0.064 food/step)
Running generation 89... complete (1.34799 seconds, 1151 steps, 48 consumed, 0.044 food/step)
Running generation 90... complete (1.15939 seconds, 833 steps, 48 consumed, 0.064 food/step)
Running generation 91... complete (1.23471 seconds, 845 steps, 48 consumed, 0.064 food/step)
Running generation 92... complete (1.17346 seconds, 860 steps, 48 consumed, 0.064 food/step)
Running generation 93... complete (1.16051 seconds, 779 steps, 48 consumed, 0.064 food/step)
Running generation 94... complete (0.820653 seconds, 488 steps, 48 consumed, 0.104 food/step)
Running generation 95... complete (1.18969 seconds, 943 steps, 48 consumed, 0.054 food/step)
Running generation 96... complete (0.91426 seconds, 545 steps, 48 consumed, 0.094 food/step)
Running generation 97... complete (1.10312 seconds, 783 steps, 48 consumed, 0.064 food/step)
Running generation 98... complete (1.14092 seconds, 739 steps, 48 consumed, 0.064 food/step)
Running generation 99... complete (1.0182 seconds, 750 steps, 48 consumed, 0.064 food/step)
Total time: 165.962s

^ Looks like it was still improving at the end there.

This more competitive environment will be perfect for adding the ability to sense nearby boids.