[crimson-list] Improve AI behavior : transport

[Sébastien Dailly]

A new patch for crimson !

AI sucks with transport ! It currently only use transport when no other way to reach the target is available ; with this patch, the AI check if a transport allow to reach the target quicker. The unit stay in the transport as long it give an advantage instead of moving to the target by its own.

There is a little level test wich show the problem.

[Jens Granseuer]

Post by Sébastien Dailly
A new patch for crimson !
AI sucks with transport ! It currently only use transport when no other
way to reach the target is available ; with this patch, the AI check if
a transport allow to reach the target quicker. The unit stay in the
transport as long it give an advantage instead of moving to the target
by its own.

Nice! The reason why the AI is currently using transports a little more
conservatively is simply that they pose a high risk. Most transports
are lightly armed (if at all) and thus can easily become deadly traps
for the units inside. Ideally, an AI would need to consider both the
time to reach its target and enemy presence when deciding whether to
use a transport or not.

Jens

[SteveB]

Hi guys, my main issues with the AI are :-

1) Not enough focus on manoeuvring to capture 'buildings' = it's usually a lot
easier to win
by capturing the enemy HQ than it is to eliminate every enemy unit - also, the
first side to capture 'neutral'
buildings (especially when stuffed with units) will typically win ...

2) If you see the AI approaching a building with Infantry, and you have an
infantry unit close by, you can 'hold back' and let the AI rush in and capture the
building first. On your turn, you send in your own unit and capture both the
building
and the AI's infantry !

The AI needs to 'support' it's Infantry better, eg by placing 'blocking' units
around the building to
stop you moving in on your next turn

[personally I believe units should fight inside the buildings to oppose enemy
trying to capture it,
however that's another story]

3) Finally, there is also a 'bug' of some sort that results in the AI 'locking
up' when it's
forces start the game loaded into sea transports ('landing craft') 'off shore'
[the map was quite large - 61x29 with AI having about 200 units (enemy about 150)
- most enemy units were 'hiding' in 'bunkers' or positioned on cliffs etc.,
however there were a few obvious 'targets' on the beach.
In this map, the AI played second and never moved anything

Cheers Steve

Post by Jens Granseuer

Post by Sébastien Dailly
A new patch for crimson !
AI sucks with transport ! It currently only use transport when no other
way to reach the target is available ; with this patch, the AI check if
a transport allow to reach the target quicker. The unit stay in the
transport as long it give an advantage instead of moving to the target
by its own.

Nice! The reason why the AI is currently using transports a little more
conservatively is simply that they pose a high risk. Most transports
are lightly armed (if at all) and thus can easily become deadly traps
for the units inside. Ideally, an AI would need to consider both the
time to reach its target and enemy presence when deciding whether to
use a transport or not.
Jens

[Sébastien Dailly]

Le 2014-04-16 11:06, SteveB a écrit :

I agree with you, there is a lot of work to do on AI.

But :

1. I do not want to break the existing game. That's why I would like a
test framework in order to check regression before great changes, and I
do not have time for that yet.

2. I do not pretend to make the AI a great conquirer : just improve some
local points on which I can work with a minimal exemple, and the two
patch I've sent do not change the AI strategy.

The points you mentions are more « strategical » than « tactical » (ie,
there is now really concertation between the units, the AI act like in a
stateless game — the AI rebuild it's own strategy for each unit) and
require more change in the game than what I'm ready for.

I'm thinking about adding virtuals maps in the game in order to help to
take a decision : dangerousity map on which each hex is mark by the
amount of dammage a unit can receive; transport map (which give
importance to bridge or small way between mountains), but I'm not ready
to start such changes alone.

But if the project gains a new dynamic, I would be happy to contribute
:)

[Sébastien Dailly]

Post by Jens Granseuer
Nice! The reason why the AI is currently using transports a little more
conservatively is simply that they pose a high risk. Most transports
are lightly armed (if at all) and thus can easily become deadly traps
for the units inside. Ideally, an AI would need to consider both the
time to reach its target and enemy presence when deciding whether to
use a transport or not.

I've solved this problem by leaving (or not using) the transport if we
are at less than 2 turns of the target.

We could check for each player unit if one of them could hit the
transport, but I think that this is a more general problem for AI
movement (and the AI is already looping over the player unit too much…).

As said in a previous mail, I would solve this with a dangerousity map
which could be computed only once in a turn. But this is not a simple
patch to do…