I am personally looking forward to taking out back of the beyond missions, just so I can stomp on in and blow away some jury rigged pierats frankenmech. Not very lucrative but fun and easy 
AI news?
- Thread starter Redsco
- Start date
-
We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
Having had the joy of playing @Amechwarrior modded version of the beta, his changes made a big difference to how the AI played, as he said, especially with reserving. Playing the modded default starter lance/pilots vs the AI with the same held a reasonable amount of challenge. Taking the overall strategic considerations into account it would have been generally verging on pyrrhic victories at times. My presumption is that HBS has taken it even further - so there should be sufficient challenge, and this should in no way be seen as a denegration of Amechwarrior's sterling work - he has given me many hours of refreshed fun with the beta files - thank you good sir!
I don't see that as denigrating at all, I barely know what I'm doing. I'm just fiddling with values and weights that I can't even see the underlying decision flow to. Given HBS has two AI devs that wrote the system and have the ability to create whole new variables, they are certainly capable of making the AI brutally effective. It's just they probably don't want to 90% of the time. I can take that Starter Lance with 2/2/2/2 pilots and wipe a 20m preset Lance with elite stock pilots without losing a unit, I don't expect the average player to be able to do this even after hours of playtime.
HBS has had two people working full time on the AI for more than a year. The AI build we saw in beta is now 8+ months old.
Indeed. I'm will be fascinating to see how effective it will be and will be and will play a large factor in whether we see strategy gamers who are not deeply invested in the IP as fans.
I expect with two AI devs working on this since the last we saw of it, the AI is likely self aware by now and already working on the expansions by itself...
It's worth noting here that at least some of her testing uses AI vs. AI matches to save time.
Although I do think the AI will be much improved for most/all of the reasons others have mentioned previously.
I found and activated a "I can't do that dave" sound bite hidden in the beta files...
Awesome.
I can't wait to discover what gems HBS has in the release version or what Easter Eggs are squirrelled away...
Thank you for that confirmation. And it is what we have been looking forward to.
A key question for me during the AI Dev Q&A went something like this, "We know the AI controls each Enemy Mech Independently, but is there a Lance-level AI that helps ensure there is rhyme, reason and supporting maneuver between each Enemy Mech?"
While there had not been such a Lance-level AI at the time, as I recall it HBS acknowledged the potential value of such an AI-construct, though indicated that it might be beyond the scope of BATTLETECH at Launch.
While there had not been such a Lance-level AI at the time, as I recall it HBS acknowledged the potential value of such an AI-construct, though indicated that it might be beyond the scope of BATTLETECH at Launch.
We looked in to this on the last miner mission video. Turns out the AI can pick which unit to move first within the same phase, something the beta AI was unable to do and followed a simple order based on the lance config, from top down. So, some kind of shared info is being passed between units choosing to go individually or one master AI is calling the shots. What might be the most likely solution is that the AI already calculates the relative threat level a unit is in and then it might poll all possible units that can move. That's one way they could simulate a "Mechcommander" without having to write a real one. The first unit to move could compare its current vulnerability (and potential for enemy destruction, Sensor Locking, etc if we want to be fancy) against other units in the same phase. If it finds one more vulnerable, it could hand off its move to the most at risk unit before it takes the time to calculate any other moves. It could be a kind of extension of how a fallen unit will flag the AI to move out of order and always be the first one to stand once that fallen unit is available. Oddly in beta, it didn't consider fallen+shutdown states and I shutdown an entire Lance, knocking over only the last unit. The AI then proceeded to power up each in order, the fallen one last.
The framework to simulate an AI commander is there, without actually having one, but it was likely not ready for beta the same way Reserve or Sensor Lock was quite ready for showtime.
@Amechwarrior, that is highly encouraging and a great waypoint along the path to something more robust. Even if that something needs wait till post-Launch.
In the fullness of time, @HBS_dlecompte (aka Dave) will hopefully be able to shed more light on many things AI.
Of all the "BATTLETECH Behind the Hare" insights that might be ahead of us, learning more about the BATTLETECH AI Development is one of the stories I am most interested in. :bow:
Heres to hoping the campaign missions arent just 14 mechs, 10 vehicles, 9 air strikes, 21 turrets, 2 drop ships and 5 turns to kill it all in....
Then again, these guys arent Russian devs...
Then again, these guys arent Russian devs...
As we saw in the miner mission, we will have missions with specific objectives we have to do and they will pop up as we complete them.
Multiple waves of enemies are very likely and hidden enemies that pop up when least expected.
Yeah, even the Miner mission IIRC included personal testimonial from a member of the Press that he split his forces and managed to trigger the defenders at both sites simultaneously....
...it was a Bad day.
...it was a Kiva-kind of day.
So did Kiva overtune the campaign to be a 10% chance to win kind of thing? Like massively underestimating the AI and making them amazing?