Autonomous Combat League

[MarvinMan]

These laser-armed swarm bots (I call them ants) are very effective against slow, armored targets. I put 3 of them (21 parts total) running completely independent code in the arena with a PressureTank, and one of them was the last bot left intact. Two of the ants got in a bit of a friendly fire incident and completely disassembled each other, but still managed to distract the PressureTanks turret, allowing the third one to get close enough to remove a wheel and do some serious chassis damage.

The surviving ant did beach itself on the top of a ramp after that, but was otherwise intact and functional.


After being unstuck, the ant went back to achieve mutual annihilation of itself and the PressureTank.


The next thing to do is tweak the mechanical design so it can deal with ramps and add a bit of coordination between multiple bots. I might have a match against the erazer too, as I suspect the result will be interesting.

UPDATE: 3x ants V erazer resulted in a decisive victory for the ants, with 2 fully functional survivors.


Does anyone know why lasers don't keep the colour they were set to? The lasers on the ants start out blue, but change to the default red after a while.

[z26]

idk, but in my case when i wanted a "black" laser (actually a dimmer white) it wouldn't keep the color so i used a cyclically activated multiplexer to refresh it, so that one potential way to deal with the issue.

[MarvinMan]

I've already got a rapidly cycling toggle for the camera system, so it should be easy to add colour I/Os for the laser.

Here's the blueprint for the laser ant (V6), with slightly improved target acquisition system. The detection threshold is set to 8 parts to prevent targeting other ants, so this version may have problems against smaller bots. https://www.dropbox.com/s/nsedzsmt2zwwy7p/bp_ant6?dl=0

Does anyone have the blueprints for the later versions of the erazer? I'd be interested to see whether it has any better luck against the ants than the original. It seems as though while the erazer needs a large open map to work well, that gives all the ants a clear line of sight for them to all lock on at the same time.

[z26]

so it seems like swarms bots are as effective as hoped, their main problem tho is being able to handle tricky terrain without being made of too many parts.

edit:  If swarm bots gets too annoying for big bots to deal with  (especially since given their low part count distinguishing them from debris is difficult) , maybe dividing the acl in two divisions, single bot and multi bots would be preferable.  Bot being defined as something that can propel itself in a controlled direction.  That would mean guided rockets would be considered bots.  Darts steered by fins might be tricky however.

If you could get the current hitpoints of a part (with the connectivity meter or something else) distinguishing between debris and targets wouldn't be as difficult.  Worst case scenario: if you target a healthy debris, instead of obsessing over it you only deplete its health to 0, then move on to another target.

Or maybe see when there is only a single input to the connectivity meter whenever that part is attached to anything at all whatsoever.  That could fulfill the same need.

[MarvinMan]

The current ant swarm bot uses the almost same navigation system as my full combat bot, so they shouldn't have a problem with tricky terrain from a software perspective. The hooks need tilting upwards 20-30 degrees (like they are on the RedEye) to let it drive up ramps, and the center continuum needs shifting slightly to alter the balance and raise the ground clearance. The first ant with the cannon was perfectly capable of dealing with ramps, but had fairly serious aiming recoil problems.

Swarm bots seem to be less effective in the tighter, more complex maps where they are more prone to getting stuck on each other and it's easier to pick them off one by one. I like to filter detected parts to only lock onto parts moving faster than a threshold velocity, which should prevent the bot locking onto piles of debris while still being able to see small bots.

The PressureTank did seem to be getting close to hitting the ants, so some of the newer bots with better turrets might be able to fend them off reasonably well. Hover bots should be relatively immune being mobbed by swarm bots, given how much adding a cannon with elevation control would add to the part count and how much recoil affects such light bots. I don't know how well swarm V swarm matches would work as you'd need some sort of IFF system (the current ant bots work entirely independently and simply ignore anything ant sized or smaller), and any coordinated attacks would likely end up with every bot trying to pile into the same space.

[z26]

I like to filter detected parts to only lock onto parts moving faster than a threshold velocity, which should prevent the bot locking onto piles of debris while still being able to see small bots.

Oh, that's good!  I'll use that.  Also, aiming indeed is an issue, but to handle recoil just increase the weight of a continuum for a couple frames when you're shooting.

[MarvinMan]

I'll have to try increasing the mass when firing (although it is quite funny seeing the bot flying backwards several body lengths or even getting some air time every time it fires a high power round).

Velocity filtering can be a bit unreliable as it takes quite a while for the calculations to complete relative to how quickly the rest of the latch works. Often, the velocity calculation will keep the latch enabled after the camera has lost sight of the part, resulting in no target part being stored in memory and the bot wandering aimlessly. Adding an extra latch directly connected to the cameras detected and activity outputs seems to fix this.

[PressureLine]

To be fair, the turret on PressureTank turns far too slow afair, which causes it's accuracy to be poor because it is actually fairly fast (fast enough to generate misses against a stationary target)

I've got plans for another plasma-turret middleweight bot  that looks to be an interesting build at least

[MarvinMan]

The turret doesn't seem to be too slow now that bots tend to follow their target, as the target bot will usually approach or follow head on. A lot of turrets seem to have difficulty hitting low targets and end up firing over the top of them (which would be center-mass hits on a more sensible sized bot), especially at close range.

Good luck with the new bot, I'm still waiting for someone to build one with multiple independent turrets.

[PressureLine]

The turret doesn't seem to be too slow now that bots tend to follow their target, as the target bot will usually approach or follow head on. A lot of turrets seem to have difficulty hitting low targets and end up firing over the top of them (which would be center-mass hits on a more sensible sized bot), especially at close range.

Thats what the shape of HailStorm was all about. I think I managed to be able to allow -20° on the cannons of Hailstorm, which allows it to shoot very close to itself without shooting itself.

Good luck with the new bot, I'm still waiting for someone to build one with multiple independent turrets.

Considering that the code will be a direct port of the HailStom code, it's more an attempt to see how different construction philosophies affect the effectiveness of the bot. I did start on one flying bot with independent turrets, but it's too big and probably wouldn't be that effective.

[MarvinMan]

I ran some big matches in an open arena, with the PressureTank, RedEye, SawTooth, HailStorm, erazer and some ants. The erazer usually went first, getting caught in the cross fire and apparently being the ants favorite target. The SawTooth usually did the mast damage, almost single-handedly winning one of the rounds. The PressureTank generally stayed out of the close range mess that the newer bots got into at the start, and was eventually immobilised by the SawTooth or one of the remaining ants. The RedEye spent most of the matches on its side at the edge of the arena, though that did let it survive one round virtually intact. The HailStorm usually got into a firefight with the PressureTank at the start and was the next to be taken out after the erazer.

Overall, hiding seems to be the best way to win, waiting for the enemy to get stuck (or be involved in a friendly fire incident for swarm bots). More usefully, a cannon for long range bombardment is good to have, but close range laser attacks are the most effective.

Regarding HailStorm V ants, accidental death was the main cause of a loss in a complex map as they seemed to have trouble locking onto each other (the ants rely on the target coming to them to some extent). In the open arena, the laser ants were as effective as usual.

[Sleeping_Owl]

Lancer (25 parts).
Uses three grenade-lances to charge the opponent.


https://drive.google.com/open?id=0ByzpYkai1-8lVTZkRFVOdnRMUHc

After a few fights I did it looks quite effective for a lightweight if it manages to get into fight. Some fights ended with it being destroyed from behind (no rear looking camera, no armor on the back) or navigation failure.

Fights:

[MarvinMan]

Cool. That looks like it'll be very effective (even if it is a wonderfully massive amount of overkill against ants). Have you tried it against any of the bigger bots?

The lances look like they make the bot rather long, especially when deployed. It'll be interesting to see how well it handles the tighter or more complex maps. It might be worth adding some elbows to fold the lances and a bit more armour so that it can take on the big bots that the grenades will be most effective against.

[z26]

unconventional design, i like it.

[Sleeping_Owl]

I tried it on heavier plasma shooters. It is still effective, but you can see it take serious damage. The main problem seems to be lack of rear camera, but it can't have more parts and stay lightweight.
It is quite bad in tight spaces or just fighting near the not smooth wall, but I don't think adding a vulnerable elbow in front of the bot will make its life easier.
If I ever decide to make it middle or heavy I would add rear camera, turret and grenade elevation, and make it less vulnerable to top-down fire (HailStorm can fire over its armor).
Not a conventional turret but something like what was on Lancer's early prototype. Lance arcs would go through armor on bot's body.