A.I Fight: 1000 Warriors vs 1000 Musketeers… (Deep Reinforcement Learning)

[@Zuzelo]

What soldier shall I add next: Trebuchet, Medic or Gunner? Or shall I retrain Musketeers and Warriors? And, since you are reading this, do not forget the comment for the KZbin Algorithm Gods!

[@Dekiel]

Medic!

[@cheezum]

Trebuchets sound interesting.

[@1z34t7]

A sniper

[@Zuzelo]

@@1z34t7 But how would it be different from the musketeer?

[@bobbillyboy16]

you should make it so warriors near othere warriors can tell them to attack. so like a radius that they can use to tell each other to head to battle or something.

[@brokenearth7079]

I think the reason for the wave strategy of the warriors is the fact that the ones in the back can't see through their allies, and so aren't aware of the army yet

[@skaimbauer5556]

Give the musketeers a better reward if they shoot an enemy from far away

[@ညေြိ့ငမ့မကတြာုကသ်မစျြြငိုချ့်ဆ]

Yea

[@Rounbug]

Meet the Sniper

[@ahmedayham1273]

But also a punishment for being too far to give a middle ground

[@zeemon9623]

Cavalry as a fast-moving unit could dethrone the musketeer from (ironically) his high horse.

[@Zuzelo]

Get Out! I like the idea tho :D

[@mrfrog0913]

​@@ZuzeloThe Idea Cavalry Pogos charge into the back lines and kill the enemys. When Cavalry Pogos in the real game see musketeers *War flashbacks*

[@topherrybicki19]

It would be interesting to see less ambling around by the backline units. Maybe you could tell each unit where the center of mass of enemies was located? Maybe you could even add a leader unit that knows where enemies are better and can make allies follow him.

[@Zuzelo]

Haha you've read my mind I am actually planning on adding a commander unit in the near future :D

[@mrfrog0913]

​@@ZuzeloI think other units should get a reward for staying close to the commander, and the commanders would increase the damage of nearby Pogos

[@HaunterV]

@@Zuzelo just a communications unit that alerts nearby units where the majority of the enemies are. i guess im talking about a scout, and then later add the commander unit that weighs the scout information and gives nearby units either the go ahead to head to enemies, or to hold the line

[@nikolozgilles]

Omg the final battle like the battle of Rorke's Drift lmao ZULUS ATTACK FIGHT BACK TO BACK SHOW THEM NO MERCY AND FIRE AT WILL KILL OR BE KILLED FACING AWAITING A HOSTILE SPEAR A NEW FRONTIER THE END IS HERE THERE'S NO SURRENDER THE LINES MUST HOLD THE STORY TOLD RORKE'S DRIFT CONTROLLED

[@Zuzelo]

FOR HONOR! FOR GLORY!

[@marcboss]

Thanks for posting!!! This is my favorite series, please make more!

[@Zuzelo]

Oh I will! Glad you like the series :)

[@j.j.maverick9252]

called it. with the big teams separated the warrior training didn’t know how to use superior numbers to charge en-masse and create a spearhead. Think pre WW1 battlefields: the “over the top lads, all together” stupidity against machineguns was a really good tactic before machineguns.

[@j.j.maverick9252]

oh and thanks for the video, very entertaining!

[@Zuzelo]

True, Normandy was not recreated, but we ain't gonna give up! I will keep experimenting with these soldiers while adding new types :D

[@nemoumbra0]

Currently the soldiers are controlled by the neural networks. I suggest adding a possibility to replace the "backend" of the soldiers with an AI that would demonstrate the desired behaviour. For instance, you'll be able to test how well the Musketeers would perform against the enemy whose brain is instructed to rush in the direction away from the wall and then start attacking (the grand offensive). This is how I imagine my own neural network tests would go.

[@Zuzelo]

Hm, that sounds interesting, I like that! Basically putting the AI in all kinds of custom made situations. I might actually do something like that in future episodes! Thanks

[@nikolozgilles]

I think they meant to program a very basic baseline AI to test/train the soldiers against, rather than each other. Something like a soldier that runs straight into the nearest enemy and attacks very dumbly

[@nemoumbra0]

@@nikolozgilles That's right! I think ZuzeloApps got it. This will be used to create custom situations.

[@winnerwannabe9868]

I think that sounds like ppo, but I could be mixing up the name or misunderstanding something.

[@zqmxq]

Idea: Shield units Shield units cannot attack but have 500 health and take 50% less damage from ranged units. It would be interesting to see if the AI makes different formations with more units

[@cheezum]

LET"S GO!!! Been waiting for this video! Keep up the great work!

[@Zuzelo]

Lets goooo!

[@simonosadchii5363]

Great video. Laughed from moshpit being introduced. I have a question about training, how collaboration between agents was encouraged? Also, many warriors were standing still for the most of the battle, maybe that's because they were trained by moshpit spawning strategy, so in a way they expect enemies to be in their observation range. Would also enjoy seeing some obstacles on the map, like a wall or elevation (small hill) and see how the strategy evolves! W

[@Zuzelo]

Glad you enjoyed it :) The main encouragement for collaboration was team rewards and punishments for winning/losing. In case a team wins, the entire team gets a reward and all opponents are punished. That is a good theory why the are standing still. My assumption was that the warriors from the backlines were not seeing any opponents (because they were only seeing their allies from the frontlines), so they were just messing around. Once the front line died, they actually are able to see the enemies and start charging. Might be a mix of both

[@nikolaipasko]

Perhaps Pogos just don’t see enemies from behind their buddies and just don’t know where to go to fight. Some ant-like mechanism could help - like a signal to neighbors: “enemy, musketeer, 12 o’clock, 200”

[@gerardjagroo]

New idea: Charge Attack! Warriors get rewarded for charging while attacking. Maybe you can finally get both side to charge into one another like real battles?

[@Zuzelo]

Ooooh you gotta check the latest video!

[@nikolozgilles]

Help I just woke up and there's been a spider hanging over my head on the wall this whole time i'm scared

[@x-ani4061]

for 5k subs your videos are pretty high quality. good, good.

[@Flemmi]

a thing you could change about friendly fire is that agents that are right next to one another dont hit each other but once there is a bit of a distance (like the lenght of one musket) they can hit one another, it feels to me the current set up rewards musketeers for being lone wolfs and not be a unit.

[@Zuzelo]

In later videos they actually form line formations!

[@skrelvthemite]

it would be cool if you added certain obstacles to the field so pogos can hide behind them for cover

[@Zuzelo]

Yeah, I will definitely try that! Not sure how efficient those will be tho

[@vikingsundlof9040]

4:16 Red

[@iuliusconnor6918]

European colonisers meet the locals: a simulation.

[@Zuzelo]

xD

[@MrTinkerbell72526]

Pogo guner

[@jazzplane5807]

Good video & Great humor

[@Zuzelo]

Glad you enjoyed it :)

[@dapperpotatoes8473]

You could potentially include melee friendly fire but diversify melee troop types, such as spears which would be much stronger for formation fighting. Or cleaving glaives/poleaxes that have great reach and massive damage potential at the cost of their formation strengths.

[@lifefindsaway7875]

Do we need to change the AI to encourage formation fighting? You mentioned that they can see enemies, can they see Allies? Do they need to be told to notice what direction their Allies/enemies are facing?

[@Zuzelo]

In latest videos there are definitely more formation attempts from the agents Agents can both see allies and enemies and I have also tried making them see the rotation in the latest video

[@Bingusgodd]

You should make 2 empires with many different soldier types then make them fight. Also add Jeff

[@Zuzelo]

That is the final goal :D Also, noo, not Jeff!

[@liamomarrah7039]

The soldiers will win (i mean warriors) 4:17

[@KacperBe_]

I think that too

[@XPAction]

Try to make a map with a trench in the middle.

[@Zuzelo]

I might actually! However, first I gotta optimize the existing environment a little :D

[@Lowgraphicsdeer]

I have watched 3 of these, it is 11:35 pm

[@Zuzelo]

3?! Those are rookie numbers, you gotta pump them up!

[@GOJOUchiha-kj8po]

where did you learn all that ?

[@Zuzelo]

Basics of programming in the university, the rest is practice :)

[@Thatkid461]

Why did bob get 14 dmg instead of 50?

[@Zuzelo]

idk, most probably stepped on a Lego piece

[@levoid9187]

I think they might have attacked at around 200 troops remaining because the amount of troops on the battlefield disturbed the pathfinding system you were using. They probably try to avoid each other right?

[@Zuzelo]

They have no path finding. They use raycasts to see the distance towards obstacles and then can decided to move/rotate in that direction

[@MrRobsn89]

Warriors should be able to walk towards the enemy when holding the shield and rush .. as it is now the stamina is used up too fast I guess so they will not use it because punishment for no stamina left etc is too high

[@yourfavoritehistorystudent1377]

Reward moving toward enemies punish moving away from enemies

[@LabdalockVelheimr]

he should've made it so that they become ragdolls when they die

[@-Nue-]

You should make blue teams blood blue so they can be seen better, "Blue Bloods"

[@Gremlin0]

R E D ive gone red every time and ill do it again. I have faith in those little guys

[@Zuzelo]

Not this time bro bro :(

[@Jesiah_Savage]

I think they don't rush because they can't see the enemies through their own troops

[@co1t_on_yt]

The last warrior: 🫡

[@michaelpelikan94]

which software is this?

[@Zuzelo]

Unity + Unity ML Agents

[@YourLightning28]

That was more like Omaha beach than anything else

[@UserK-46811]

have the camera center on a player , just lkie UEBS2

[@Hahatomato]

Do you know what’s bigger than bob’s hp? His c

[@thej-ster8826]

👍

[@lasse9272]

Algorithm 😢😮😂🎉❤😅😂😊🎉😮😂💀💀🎂🎂🎂🥰🇦🇹🇦🇹🇦🇹🍌🍌😩😩😩🥹⚧️🥳😍😥🤯

[@MrTinkerbell72526]

Bob

[@skaimbauer5556]

Make a sniper tower who can spot enemys and send soldirs to the enemy

[@Zuzelo]

Oh that sounds similar to the commander that I will add in future!

[@colegilbert673]

i expected musketeers to win because gun

[@Zuzelo]

same xD

[@Zoro12386]

Pogo red

[@Mr.Ice777]

BLUE

[@mrchief975]

go musketeer

[@mrchief975]

YEAAAA

[@JK-eu3qf]

Musketeers

[@ashurashur9368]

Team red and Medic

[@1z34t7]

Red Warriors pogo team will win

[@Zuzelo]

:'(

[@AirConditioninUnit]

Add suicide bombers and try to teach them to get multiple kills with one attack.

[@Zuzelo]

Why do I like this idea so much?!

[@fulookin6701]

European vs British

[@Zuzelo]

vs Bri ish*

[@sachas6316]

I think blue is gonna win

[@Zuzelo]

Noice!

[@Мопс_001]

The whole idea with the battlefield went far less exciting than I thought it would be. The training just seemed wrong altogether. Warriors not understanding how to fight properly, muskets not using any strategy except to stand and shoot and etc. To be honest, there is barely any strategy here, only very rudimentary elements of fighting. Would be more spectacular to watch with proper training for each single unit, perhaps on a much smaller board so they understand the basics of fighting. You don't train a bunch of robots in one shot. You train one good robot which you make a bunch of. I'm by no mean a programmer but I don't think just making the training extremely long would be efficient with such a poor training

[@sitaandini8584]

No the musketeer atk damage is nerfed to three hits

[@poetrywithbennit]

What happened to MVP system

[@josephinehull8633]

RED

[@randomanticswithbobandcrai584]

Hi im bob.

[@Zuzelo]

oh no, he is here! RUUUUN

[@stephenmensah3142]

for zulelu oikip

[@user-px4vs2yh7l]

Red worriors pogo will win

[@MrTinkerbell72526]

Sponsbob

[@person4579]

blue warriro wilkl win

[@jvagell8699]

red

[@mboop127]

humiliating yourself by including Elon Musk in a video