Making the Bots "Better"?

[johnboyjjb]

When playing ATB, I have my units sorted into lances, companies, and battalions. Force view shows that quite nicely. But the OpFor is just an unorganized horde. I will sometimes control a lance and set the other two lances in my company into bot control, each with a different bot controlling them.

Is there a way to break the OpFor into lances in the force view, break them out into unique-ish deployment zones, and have them on teams but act as separate entities?

Should there be an RFE to have every bot unit have a specific hierarchy of precedence for the herding setting?

Should there be a toggle at the end of turn screen to boot (or even better - reconfigure) all the bots (or specific bots)? This would mean that I as the commander could assign a target for a lance under me, tell that lance how tightly to stick together on the board, and when either the target is removed or the lance decreases in size I could re-prioritize the targeting and herding instructions. This could also mean that the bot could be told to herd around a specific unit, defending a designated target.

Thoughts?

[Lanceman]

Is there a way to break the OpFor into lances in the force view, break them out into unique-ish deployment zones, and have them on teams but act as separate entities?

Not convenient for AtB since the teams are preconfigured, but in normal matches you can add several bots, each assign them different behaviors and units, and stick them on the same team. And with the new deployment options added in the last couple of versions you can tweak starting placement a lot. You still won't get any kind of cohesion or "tactics", that's just the nature of the AI, but you can mix things up a bit.

[Hammer]

We have lots of players that once MegaMek launches from MekHQ. They configure each princess to have a specific role. But ultimately Princess is limited by her inability to think turn over turn. She is strictly limited to min-max for the turn she is in.

So, until someone comes along and codes Skynet into MegaMek there are limited options.

[BATTLEMASTER]

It makes me wonder what the computing requirements would be for MegaMek to evaluate the OpFor's possible moves 6 turns ahead for each unit, and re-evaluate those moves after each move in the turn.  The bot would move its units to respond accordingly to what it thinks is going to happen.

This makes me wish I was an AI pro!

[pokefan548]

Accounting for players using optional rules, large mapsheets, and potentially quite large armies, even if it ran much more efficiently than Princess does it would still take ages to plan that far ahead, and by the end of it the Turn 6 forecast would probably be so vague, with so many possible outcomes, that the bot would honestly be a more effective fighter if you threw that data out than if you were to try and make it worry about every possible state of every unit on any hex they can reach in the span of a full in-game minute.

[johnboyjjb]

I was really just interested in getting the bot to follow a unit and the UI allow for more prompted convenient bot kicking.

[dgorsman]

The herd setting in Princess considers all units on it's side, so there isn't really a way to get it to follow an individual unit.  You could try plugging in a friendly unit into the Preferred Targets list and set that Princesses preferred engagement range closer to Berzerker, and see how that goes.  You can adjust the herding and other settings of friendly bot players on the fly using the command line, plus assign additional preferred targets.

You should be able to split up OpFor units into multiple Princess controlled players beyond what is passed by MekHQ, through the lobby in MegaMek.  However, I am uncertain how this will affect the scenario resolution dialog for things like "Preserve X% of forces" requirements.  For example, any units which are deleted from the MegaMek lobby have no check box present in the scenario resolution dialog.  That will require some structured experimenting to see what the results are.

It makes me wonder what the computing requirements would be for MegaMek to evaluate the OpFor's possible moves 6 turns ahead for each unit, and re-evaluate those moves after each move in the turn.  The bot would move its units to respond accordingly to what it thinks is going to happen.

I think it would be more productive for Princess to engage in forward thinking in a more strategic fashion with units that are not directly engaged.  In other words, lighter units with a Scout or Striker role would be sent out to places on the map that cannot be seen, or to investigate units that are sensor contacts; while heavier units with Juggernaut and Brawler units are sent to a map location that is considered important such as a crossroads, bridge, or hilltop.  The best we could hope for, I think, for predictive fighting would be an on-the-fly tracking of the order of movement so Princess would do it's calculations based on expecting a certain unit to be moving based on how the human player has been playing.  The same technique would also be useful for dynamically adjusting the suggested order of movement for the human players, so it isn't constrained to the same (and frequently annoyingly wrong) order.

[johnboyjjb]

Breaking up the the OpFor into different bots does NOT seem to have an obvious negative effect on resolution. I've done it bunches of times.

If a scout lance was attached to a scout bot and that bot could be told to go to a specific hex to investigate, that would be similar to having the scouting of sensor blips.
Likewise, if your juggernauts and brawlers could be told to go to a specific hex and sit there until loosed would be the same as guard duty.

Of course, if you had that, somebody would request having the bot patrol between a series of hexes.