UPDATE [OCTOBER 10TH]: We are starting to lock in on our release candidate for next week, but we wanted to check in with a quick update confirming what next week looks like and what we're wanting to see from this weekends tests. As it stands today, we're seeing all the right signs that the Hotfix is effective at restoring Lizardmen and Tomb Kings AI to a better state. They're recruiting armies, challenging nearby factions, and interacting with campaigns in line with our expectations. Similarly we've observed Chaos Dwarfs are now recruiting elite dwarf units, though we will call out that many of their army comps that we've reviewed still show them filling their armies with Hobgoblin/Orc Laborer units. Chorfs aren't a focus of this Hotfix, this is all about LM and TK, but it is something that we'll be wanting to follow up on after we've rolled out 6.3.2. We'll also share that LL's like Golgfag and Changeling are something that we're discussing internally. We're assessing how we want to approach redefining some of these factions ability to influence Campaigns when under AI control. Naturally any bugs these factions exhibit will be resolved as a priority, but we will take much of the data that we've gathered these past few weeks (including player feedback across our channels) and use it as a starting point to map out future changes we may look to bring here. Back to topic - this weekend, we're continuing to run lengthy campaign playthrough tests across our development teams, bolstered by a further run of Autotests to help provide us with masses of baseline data. On Monday our teams will meet to review the results of these tests and start to make determinations on when exactly we'll publish the Hotfix next week. If we continue to see more of the positive results that we're already seeing for Lizardmen and Tomb Kings, then we're all speed ahead for publishing it to LIVE as promised. If we feel that there's any risk of the issue persisting, or any new issues starting to show in our tests, we'll carefully assess how we proceed from there. Our objective is to ensure we get this right, and not to add more Hotfixes into the mix, which thankfully isn't something we're expecting that we'd need to do based on the results we're seeing so far. Either way, we'll continue to keep you updated as we push on into next week. |
Hi all,
We’ve been investigating the issues affecting Campaign AI since the release of 6.3.1 and are working hard to restore it back to a better standard. I want to give you an update on why these issues have occurred, and where we are with our progress on fixing them.
A Hotfix is in development, currently its singular focus is on Campaign AI Unit Cap Recruitment fixes and the bad AI behavior that stems from this issue. Hotfix 6.3.2 is presently scheduled for release next week. It entered the first phases of our testing earlier today and is showing positive signs of providing an improved experience to the game.
Over the last few weeks when responding to questions on this topic, we had originally planned to publish the fix as part of 7.0. Now that the investigations have been completed and we are almost there with a fix, it makes sense to decouple it from 7.0 to get it out as quickly as we can.
The issues with Campaign AI are unfortunately complex to resolve. We haven't been able to deliver an immediate fix as we needed to conduct some very thorough investigations into the root causes, but we are working as quickly as the complexity allows. We’ll go into that complexity below for those who want the detail, but if you’re just looking for the headline: we’re on it.
Hotfix 6.3.2 aims to address the recent problem that we’ve seen where factions aren’t recruiting units into their Armies, and the idle behavior that’s stemmed from that. This issue with recruitment has been a highly visible problem since the release of Update 6.3, but this issue wasn’t caused by 6.3 itself. We’ve discovered that this issue has been present in the game prior to this update and affects factions where the AI is tasked with managing Unit Caps. We have found that the AI was building lists of units to recruit without taking caps into consideration, resulting in recruitment failing to occur and stalling the AI decision making process.
These issues have been compounded by changes that we made to the different resources that are required to recruit units by the Lizardmen and Tomb Kings, and why you may have also seen issues like this happen with factions that use pooled resources (like Spawning Sequence, Meat, Oathgold, or Skulls) to recruit. Lizardmen and Tomb Kings will still face an uphill struggle in their campaigns when managed by the AI (they have challenging starting locations which regularly sees them defeated fairly early on) but they shouldn’t be fighting with both hands tied behind their back.
For a deeper look at what causes these issues, here’s Lead Technical Designer, Radoslav Borisov with detail about how our AI is currently behaving.
The way our AI handles unit recruitment occurs over several distinct steps.
One of those steps is selecting a list of units to recruit into a specific force, with the goal of acquiring the necessary strength to perform a task.
When the AI is mapping out its shopping list of units that it wants to recruit, unit caps are not currently taken into consideration in a proper fashion. This results in the recruitment action failing as soon as a unit's cap is exceeded.
Recruitment action failure then occurs. Should the AI decide that its next task should be to attack a settlement, it’s generating a recruitment action and then expecting an increase in force strength to a level where it feels the settlement garrison that it's targeted for attack can be defeated. Without the unit recruitment task’s successful resolution, it holds up the subsequent task and leaves the AI in a paralyzed state. It’s a cascade of failures that result in certain factions failing to complete any aggressive actions whatsoever.
In order to be efficient in how resources are allocated and spent, the AI relies heavily on several beliefs around the current state of the game world.
Some examples of these assumptions are things like:
Cheapest unit that can be recruited anywhere in the faction
Strongest unit that can be recruited anywhere
Most cost-efficient unit (best cost to strength ratio)
Estimated number of turns to reach the recruitment location of the strongest/cheaper unit
Any mistakes when constructing these assumptions has been found to lead to a catastrophic failure in many of our AI systems.
If the AI believes a unit is free and provides any meaningful strength increase it will not allocate any resources to buy units – it wrongly believes it can fill its armies with powerful units for free and so will pursue that option and trigger a failure cascade.
If the cost of a unit is not properly evaluated, the AI finds itself in situations where it has budgeted some money that ultimately ends up being insufficient, causing overspending and running them into an irrecoverable, or very slow to pay off debt.
This is where pooled resources come into play – the AI’s ability to understand, plan and budget pooled resources is not ideal. Lately, AI has not been factoring in pooled resource to its costs properly, leading to incorrect beliefs about what it can and cannot afford, resulting in action failure.
These past weeks of investigation have shown to us that the majority of our internal systems were unprepared for actions that ostensibly could not fail, to fail. The cascading effect led to all sorts of problems – the AI couldn’t change stances properly, attacking on the campaign failed, recruitment failed, laying siege failed, and so on.
We’ve identified and resolved the leading causes for such failures, but it’s very likely there are other cases we are not yet aware of just yet. Resolving the immediately known causes of this problem is helping us to remove any denser levels of fog that may be obscuring other possible causes, and as they become known to us, we’ll resolve those too.
As it stands today, there are around 200 different pooled resources in the game and they are used in a large variety of ways. For us to be absolutely certain that everything is properly accounted for is a daunting task, but we will continue working on identifying and resolving any issues in the future, and will not deploy this Hotfix without careful monitoring of the effect it has, and will continue to stay committed to bringing more improvements as necessary.
— Radoslav Borisov // Lead Technical Designer
News on Tides of Torment will take a back seat until we’ve resolved this issue. We’re looking forward to giving you your first complete look at this next DLC, but fixing this comes first.
In closing, please accept our apologies for the experience that you’re currently having with the game. I’ll be active across our different community spaces helping to keep you informed on our progress as we move towards the release of Hotfix 6.3.2.
— Freeman // Head of Community
For the latest news and announcements, don't forget to visit our official social channels.