Development Update #9 - AI

[p]
In this episode we wanna spotlight the latest ai changes mainly driven by our new AI programmer, better known as Solarint. You may know him from the AI mod for SP-Tarkov called “SAIN”.

Before we start, we want to get ahead of any possible confusion around the use of the term “AI”. When we talk about AI or bots, this has nothing to do with LLMs or any of the hype around “AI”. This is a normal game system inspired by decades of work from other developers.

The point of this dev-log is to list the major new mechanics being introduced and explain the reasoning and goals for each system.

So, let's talk about what we’ve been working on for the past few months.

In 1.3 we will introduce the first version of RedFor - an AI system for all bots in Incursion Red River. Shoutout to our discord community who suggested the fantastic name!

[/p][h2]Overview[/h2][p]
AI behavior has been completely rebuilt. It will feel fundamentally different to the enemies you will encounter in 1.2.

We are aiming for grounded and realistic behavior. Bots won't be doing jumpshots or anything that a person in their situation wouldn’t plausibly do.

This first version is designed to be a solid framework to build off in future updates. That means it is not feature complete and some behavior is placeholder until the systems can be made to support it as intended. Even so, we think it will be a very compelling upgrade compared to 1.2.
Just like in 1.2, RedFor is using a version of Goal Oriented Action Planning or “GOAP” called “Hierarchical Task Networks”.

This is certainly not a comprehensive list of everything new, but should give you a good idea of what we are designing going forward.

All the clips shown are recorded in editor on a test map, unscripted, and mostly unedited. Everything is actively in development and there are several issues currently being worked on.

[/p][h2]Movement[/h2][p]
The way bots move around in the world has been something of particular importance. We want bots to have a sense of physicality - like they exist in the game world. This will be an ongoing process, but for 1.3 we have rebalanced the speed at which bots move in all situations, and implemented inertia in their movement along with many smaller changes.

We want them to mess up sometimes as long as their mistakes feel like human mistakes rather than buggy ai. But of course - It's a fine line here and this is going to be something I’m closely watching as play testing begins. Designing bots to fail in interesting ways is a lot more difficult compared to having them execute everything perfectly.

[previewyoutube][/previewyoutube]

[/p][h2]Dynamic Lean[/h2][p]
Bots can lean around any corner at any time, and will try to minimize how much of their body is exposed to an enemy.

[previewyoutube]**Video**[/previewyoutube]

[/p][h2]Difficulty: Challenging but Fair[/h2][p]
The goal is to make every enemy feel like a threat and require respect, even for experienced FPS players. If things are too easy, it cheapens the experience.

So much has changed that it's difficult to give an exact measure of how much more difficult it will be compared to 1.2. In some ways, it may be easier, and in others, it may be harder.

At the same time, accessibility for players of all skill levels is important. We believe in allowing players to modify the experience in the way they seem fit. At risk of sounding corny - we want to “Empower the Player” to play the exact game they want, while also providing a solid default experience for those who’d rather just play.

So if you either want to chill with friends, or if you want an absurdly hardcore experience where death comes quickly, we want to support it.

[/p][h2]Stealth[/h2][p]
To us, stealth is not lost when combat pops off. It's something you should always be thinking about to reposition and put yourself somewhere that your enemy doesn’t expect to give you an advantage.
Currently we have a few key features related to this:

• All bot senses have an enforced reaction time before they are allowed to perceive it.
  
• Bot sound occlusion simulation. Bots hear sounds at a reduced range depending on what's between them and the sound source.
  
• Bots who are un-alerted will have reduced peripheral vision and perception range for all their senses.
  
• Bots who are about to see you will turn towards you and look before fully spotting you.
  
• Vision speed depends on how far you are from where they expect you to be, if you flank they will be slower to see and react to you, while if you are where they expect, they will see and react to you faster.
  
• Poor weather conditions will be reflected in how well bots can sense you. For example, rain will dull their hearing and vision depending on how intense it is.
  

[/p][h2]Focus[/h2][p]
This system gives bots the spatial awareness to understand where they should be pointing their weapon at all times.
This has 3 main purposes:

• It completely solves the classic AI issue of staring into walls because they saw or heard an enemy behind that wall.
  
• If their enemy happens to appear where they expect them to appear, they can react faster and in a more believable way.
  
• It makes flanking more effective for the player in a believable way.
  

[previewyoutube][/previewyoutube]

[/p][h2]AI - Friendly Fire[/h2][p]
There are several new checks to make sure bots don’t gun down their friends.
While not perfect, it does make it much rarer than before. But you should expect several improvements to this in the future.



[/p][h2]Suppression[/h2][p]
Both players and AI will have the new Suppression Receiver System, which is able to identify when bullets are flying past their head and have suppression build up over time.
The amount of suppression is scaled based on the caliber of bullet, so 9mm will affect bots much less than a 7.62 projectile. While we have new visual effects for the player as well, let's focus on the AI.

[h3]AI Under Suppression[/h3]
Suppression against AI applies a debuff to their accuracy and perception, the more you shoot at them, the stronger the debuff. This builds and decays at the same rate as the player - letting you intuitively understand what counts as suppression, and how long you have before they start to recover. The system treats all characters in the world the same.

[h3]AI Suppressing Player[/h3]
They are not limited to only suppressing enemies in a scripted way. Suppression can vary from firing a few shots at where you just ran into cover before they run to cover as well, or full on barrages of bullets from multiple bots as they coordinate bounding overwatch (work in progress!). They can decide to suppress any target at any time based on how long it's been since they perceived you in some way. So they can manoeuvre to flank 1 target while suppressing another at any time.

They try to manage the ammo in their magazine to always keep enough ammo in reserve depending on the situation. For example, if they are being shot at and caught out of cover, they will allow themselves to use more of the ammo for suppression, but if they are in a safe position, they will be more trigger disciplined.

[previewyoutube][/previewyoutube]

[/p][h2]Flashlight and Laser Detection[/h2][p]
Bots can now see you shining your laser or flashlight around in the environment. That means that being careless about light discipline will give you away and cause bots to search for where they think it came from.
This system is something that is going to see major improvements and overhauls as we continue working on RedFor.

[previewyoutube][/previewyoutube]

[/p][h2]Customization[/h2][p]
The previously hidden AI customisation menu under Gameplay settings has been moved to its own tab in the settings menu with this update. In the future we will continue to expand on what you can customize including ways to share your settings easily.
I come from a modding background, and want to support that by designing systems in a way that modding is as easy as possible, and allowing you to customize the game from the settings without the need for mods.



[/p][h2]Performance[/h2][p]
Redfor is designed to make use of multithreading as much as possible. We are still in the process of optimizing, and don’t know yet how it’ll run on every system. With that being said, early performance profiling has been promising, and it should overall run better than before when playing singleplayer or hosting a server in coop - especially when there are large numbers of bots active and fighting. We are always looking to improve our systems, and we need the bots to run great for when we start adding AI vs AI fighting in the map in the future, as that will be the true test.

[previewyoutube][/previewyoutube]

[/p][h2]Feedback[/h2][p]
Feedback is fundamental to game design and we want to hear what you think - both positive and negative. While we won't promise to implement every suggestion you have, we will always listen to feedback.
As 1.3 begins play testing we will be closely watching for issues or looking to make changes based on your experience. We want to make something that YOU love, not just us.

This is not a complete list of everything new. We’ll have a more complete changelog when the update is released.

Also, if you want to know more about AI programming and what tools we use, feel free to reach out to Solarint on our discord, he will be eager to answer any question about AI!

[/p][h2]OUTLOOK[/h2][p]
We are now ramping up to the next big update for Incursion, which means we will also start testing soon! Should you wish to participate in playtesting when the time comes, be sure to join our Official Discord Server. Where you can assign yourself the playtester role, and give feedback over in the respective channels.

Additionally, if you are interested in picking up the game, the Steam winter sale just started and Incursion has a 20% discount!