Intel Report #025: Road to V1035 (January 2025)
Hey there! We hope 2025 has been treating you well so far.
Over here on the GROUND BRANCH team, we’ve been hard at work pushing towards the official release of update V1035.
As usual, we would like to thank you — our amazing community — for your patience and support while we get on with this. There are still a few gremlins stalking our code base, and a few things that turned out to be incredibly… not straightforward when we got around to implementing them, but we are working as hard as we can to get things fixed. This has taken longer than any of us would have liked but, on the plus side, it also gives us a chance to continue to stuff cool new things into V1035.
We look forward to share the update with you when it’s ready. Don’t worry — you’ll definitely be invited to join the fun before that happens, once it’s ready for public testing. In the meantime, we’ve put together a fresh batch of previews and deep dives to keep you entertained and in the loop.
One notable update we've worked on over the last couple of months is the operator customization, with a special focus on the Customize Item screen layout.
Whereas in V1034 and earlier we had the rather unwieldy nested categories, the new layout looks something like this:
[previewyoutube][/previewyoutube]
As you can see, attachments are now displayed in a larger size and organized into category checkboxes that are always in view. Multiple categories can be checked with a right-click, in case you'd like to (for example) see all sight options. There's also a search bar, which will be especially useful for the Steam Workshop fans out there who like to stock up on mods.
When customizing weapons, attachments are laid horizontally in the bottom of the screen in up to 2 rows of 8 attachments each. That is so longer weapons can take up more of the screen without being as obscured by menus. However, when customizing headgear, platforms, and belts — which tend to be taller or do not require as much horizontal space — attachments are laid vertically in the right corner of the screen.
Vest platform and belt customization will now also display your loadout summary in the bottom-left, reducing guesswork and back-and-forth. (Please note that there's a little bug in the video wherein the loadout summary displays the name of the pouches rather than their content!)
One very welcome change for weapon attachments is that sights have finally been broken down into multiple categories, reducing clutter and making everyone's life much easier:
Attachments will also finally support skin selection, removing the need for duplicate items with different skins — such as "Suppressor (Black)" and "Suppressor (FDE)" — being listed as separate items. As the video shows, we stuck a few new FDE skins in there for you. Lastly, you may have noticed a little more of an atmosphere going on, thanks to some lighting tweaks and the addition of some particle effects.
Everything is subject to change (and your feedback), but it already feels much easier to use and navigate.
This month, programmer and frequent dev blog contributor Matt "Fatmarrow" Farrow takes a deep dive on how AI patrols work in missions. One reason is that V1035 will improve on this system — which we'll explain in detail below — to provide more wide-ranging and unpredictable AI patrols. Mission modders may find this of particular interest.
Figure 1 shows a collection of patrol nodes, which are placed and linked together in the Mission Editor (see section 2.6.3 of the Mission Editor page of the GROUND BRANCH Modding Wiki for more details on how to set those up). These define arbitrary paths for patrol AI to follow, and may contain junctions, loops and dead ends. During a round, patrol AI initially go to the nearest patrol node, and then pick a new destination patrol node every time they reach their current target.
Figure 2 shows the historic method of picking an exit link from a patrol node for junction patrol nodes (that is, a node having more than 2 links). The link which is the most "forward"-facing (that is, requires the minimum deviation from the current path) is always chosen. This prioritizes avoiding sharp turns and possibly unnatural route choices when patrol AI is navigating a mesh of patrol nodes.
Figure 3 shows how, in practice, the historic link selection method can result in undesirable patrol paths. In this case, patrol AI who are on either Path A or Path B in Figure 1 will continue to follow that path indefinitely, because the most "forward" link at junctions will always be within the same path (in this example at least). It could be that the patrol AI will start on Path C in Figure 1, but after the first junction, they will get into an infinite loop as before.
Figure 4 shows a new method that we are adopting in V1035 for selecting onward links at patrol node junctions. Links are divided into "good" (anything that is remotely in a forward direction, e.g. up to a 90-degree turn) and "bad", and also "used" and "unused". Each patrol node keeps a record of which links are recently used; when all the links are used, the list is reset. A link is deemed used if it has been used either to enter or to exit a node. When a patrol AI enters a junction node, the entry link is removed from consideration, and the remaining links are categorized into "good unused", "bad unused", "good used" and "bad used". The first non-empty group in that descending order of priority is selected, and a random link is picked from that group. In Figure 4, there are shown two "good unused" links, so that group would be picked and the exit link selected randomly from it.
Figure 5 shows a more complex version of the method shown in Figure 4, in which 30% of the time, "good used" links are prioritized over "good unused" links, but otherwise the priorities are the same, with "bad unused" links taking precedence over "good used" and "bad used" links, respectively. The effect of this is just to muddy up the probabilities a little and to avoid the patrol AI getting stuck in different and exciting new infinite loops based on the new algorithm. The 70% chance of picking a "good unused" link is shared between the two links in question, leading to a 35% probability in following either. There is a 30% chance of picking the sole "good used" link.
Figure 6 shows the same setup as Figure 5, but with all of the forward links being marked as "good used". This means that the next group in the list is "bad unused", with the method normally prioritizing "unused" over "good". The single "bad unused" link therefore has a 70% chance of being selected. The "good used" links collectively have a 30% chance of being selected, as explained above. The precise probabilities are not important, but what is important is how the probability of links changes as the AI moves through the patrol node network.
Figure 7 shows the new method of selecting patrol node links in practice. The AI may follow loops, sometimes more than once, but what is now prioritized is travelling over the full extent of the patrol node network. After the first couple of junctions, the location of the patrol AI should be relatively unpredictable, at least compared to the old version. Also, the patrol node junctions are rarely as complicated as those shown in Figures 5 to 7; normally, junctions have only 3 links.
Various parameters of this method can be varied, and it is quite likely that we will be refining those during testing, so take the figures shown here with a pinch of salt. We hope that mission makers have not been too frustrated by the patrol logic in previous versions, and will benefit from this rejig of the system. We are making one or two further changes to the AI patrols that you may notice when you start playing V1035, but we will let you discover those for yourself.
And that completes our deep dive into patrol nodes!
[previewyoutube][/previewyoutube]
And here's a little co-op footage to show off a little more of new map Ranch and its audio ambiance, as well as the G36K in action:
[previewyoutube][/previewyoutube]
Environmental artist Will has taken a break from the 747 remake to do an optimization and visual pass on maps.
Although subtle in some ways, this iteration greatly improves the look and atmosphere of the maps by carefully tweaking the lighting, color-grading materials and textures, and adjusting the fog intensity and color. Vistas (the distant parts of maps that are not accessible) were also upgraded here and there.
Let's take a look at some before (left) vs. after (right) shots:
As you can see, colors are more consistent (with fewer instances of bright colors standing out), and the fog adds a greater sense of depth — making it easier to judge different planes of the scene (near vs. far) and "softening" the more distant parts of the map which are naturally rendered in less detail.
Will has also given Small Town a little facelift, improving the texturing and adding details and props. Here are some more comparison shots:
For your appreciation, media wizard Rangda has put together a few gifs of the LVAW, G36K and AUG A3 being fired and reloaded. That's it — just a little eye candy:
[hr][/hr]
And that's all we have for you today.
As mentioned earlier, there is more going on in V1035 than we're showing here. Some of the other stuff we've already shown you in past intel reports (be sure to check them out if you haven't already), and there are a few other things that we're choosing to not reveal before the actual release, because we feel it's important to keep a sense of mystery in our relationship.
Stay tuned, and we’ll see you again soon!
Over here on the GROUND BRANCH team, we’ve been hard at work pushing towards the official release of update V1035.
As usual, we would like to thank you — our amazing community — for your patience and support while we get on with this. There are still a few gremlins stalking our code base, and a few things that turned out to be incredibly… not straightforward when we got around to implementing them, but we are working as hard as we can to get things fixed. This has taken longer than any of us would have liked but, on the plus side, it also gives us a chance to continue to stuff cool new things into V1035.
We look forward to share the update with you when it’s ready. Don’t worry — you’ll definitely be invited to join the fun before that happens, once it’s ready for public testing. In the meantime, we’ve put together a fresh batch of previews and deep dives to keep you entertained and in the loop.
💡 If you missed our Intel Report #024 for November, go read it!
💡 Keep in mind this report is not an exhaustive list of V1035 features. For an overview of expected changes and additions, refer to our roadmap.
Updated customization UI
One notable update we've worked on over the last couple of months is the operator customization, with a special focus on the Customize Item screen layout.
Whereas in V1034 and earlier we had the rather unwieldy nested categories, the new layout looks something like this:
[previewyoutube][/previewyoutube]
As you can see, attachments are now displayed in a larger size and organized into category checkboxes that are always in view. Multiple categories can be checked with a right-click, in case you'd like to (for example) see all sight options. There's also a search bar, which will be especially useful for the Steam Workshop fans out there who like to stock up on mods.
💡 This new layout integrates with the new tag system (briefly mentioned in our last dev blog), wherein items can fit into multiple UI categories. This applies particularly to main items, as shown in the Primary Weapon selection. Note that the "Suppressed" category, which groups weapons that are integrally suppressed, includes weapons from various other categories. Similarly, some weapons fit both the "Precision Rifle" and "Battle Rifle" category, and so on.
When customizing weapons, attachments are laid horizontally in the bottom of the screen in up to 2 rows of 8 attachments each. That is so longer weapons can take up more of the screen without being as obscured by menus. However, when customizing headgear, platforms, and belts — which tend to be taller or do not require as much horizontal space — attachments are laid vertically in the right corner of the screen.
Vest platform and belt customization will now also display your loadout summary in the bottom-left, reducing guesswork and back-and-forth. (Please note that there's a little bug in the video wherein the loadout summary displays the name of the pouches rather than their content!)
One very welcome change for weapon attachments is that sights have finally been broken down into multiple categories, reducing clutter and making everyone's life much easier:
- Iron Sights (non-optical sights)
- Red Dot (reflex and holographic 1x optics)
- Scopes (magnified optics)
- Accessory (magnifiers, image intensifiers and whatever else may fit in the future)
Attachments will also finally support skin selection, removing the need for duplicate items with different skins — such as "Suppressor (Black)" and "Suppressor (FDE)" — being listed as separate items. As the video shows, we stuck a few new FDE skins in there for you. Lastly, you may have noticed a little more of an atmosphere going on, thanks to some lighting tweaks and the addition of some particle effects.
💡 One major element still missing from the customization UI is detailed information about each item — some kind of infobox so players don't have to guess what every item is and does. Rest assured that will be coming in due time!
💡 An update to the click-and-drag mechanic is not planned for this update due to other priorities, but we're aware it needs a revision too.
Everything is subject to change (and your feedback), but it already feels much easier to use and navigate.
AI patrols
This month, programmer and frequent dev blog contributor Matt "Fatmarrow" Farrow takes a deep dive on how AI patrols work in missions. One reason is that V1035 will improve on this system — which we'll explain in detail below — to provide more wide-ranging and unpredictable AI patrols. Mission modders may find this of particular interest.
[h3]⚠️ NERD ALERT[/h3]
This deep dive is for the nerds. If you're not a nerd, what are you even doing here? you may want to skip this section (and go watch sports or whatever).
Figure 1 shows a collection of patrol nodes, which are placed and linked together in the Mission Editor (see section 2.6.3 of the Mission Editor page of the GROUND BRANCH Modding Wiki for more details on how to set those up). These define arbitrary paths for patrol AI to follow, and may contain junctions, loops and dead ends. During a round, patrol AI initially go to the nearest patrol node, and then pick a new destination patrol node every time they reach their current target.
Figure 2 shows the historic method of picking an exit link from a patrol node for junction patrol nodes (that is, a node having more than 2 links). The link which is the most "forward"-facing (that is, requires the minimum deviation from the current path) is always chosen. This prioritizes avoiding sharp turns and possibly unnatural route choices when patrol AI is navigating a mesh of patrol nodes.
Figure 3 shows how, in practice, the historic link selection method can result in undesirable patrol paths. In this case, patrol AI who are on either Path A or Path B in Figure 1 will continue to follow that path indefinitely, because the most "forward" link at junctions will always be within the same path (in this example at least). It could be that the patrol AI will start on Path C in Figure 1, but after the first junction, they will get into an infinite loop as before.
Figure 4 shows a new method that we are adopting in V1035 for selecting onward links at patrol node junctions. Links are divided into "good" (anything that is remotely in a forward direction, e.g. up to a 90-degree turn) and "bad", and also "used" and "unused". Each patrol node keeps a record of which links are recently used; when all the links are used, the list is reset. A link is deemed used if it has been used either to enter or to exit a node. When a patrol AI enters a junction node, the entry link is removed from consideration, and the remaining links are categorized into "good unused", "bad unused", "good used" and "bad used". The first non-empty group in that descending order of priority is selected, and a random link is picked from that group. In Figure 4, there are shown two "good unused" links, so that group would be picked and the exit link selected randomly from it.
Figure 5 shows a more complex version of the method shown in Figure 4, in which 30% of the time, "good used" links are prioritized over "good unused" links, but otherwise the priorities are the same, with "bad unused" links taking precedence over "good used" and "bad used" links, respectively. The effect of this is just to muddy up the probabilities a little and to avoid the patrol AI getting stuck in different and exciting new infinite loops based on the new algorithm. The 70% chance of picking a "good unused" link is shared between the two links in question, leading to a 35% probability in following either. There is a 30% chance of picking the sole "good used" link.
Figure 6 shows the same setup as Figure 5, but with all of the forward links being marked as "good used". This means that the next group in the list is "bad unused", with the method normally prioritizing "unused" over "good". The single "bad unused" link therefore has a 70% chance of being selected. The "good used" links collectively have a 30% chance of being selected, as explained above. The precise probabilities are not important, but what is important is how the probability of links changes as the AI moves through the patrol node network.
Figure 7 shows the new method of selecting patrol node links in practice. The AI may follow loops, sometimes more than once, but what is now prioritized is travelling over the full extent of the patrol node network. After the first couple of junctions, the location of the patrol AI should be relatively unpredictable, at least compared to the old version. Also, the patrol node junctions are rarely as complicated as those shown in Figures 5 to 7; normally, junctions have only 3 links.
Various parameters of this method can be varied, and it is quite likely that we will be refining those during testing, so take the figures shown here with a pinch of salt. We hope that mission makers have not been too frustrated by the patrol logic in previous versions, and will benefit from this rejig of the system. We are making one or two further changes to the AI patrols that you may notice when you start playing V1035, but we will let you discover those for yourself.
💡 Pro tip: if you want to vary your patrol AI behavior from the method described here, one way is to incorporate "one-way" links into or out of junction nodes (manually delete links in the patrol node actors in the Mission Editor). Proceed with extreme caution!
And that completes our deep dive into patrol nodes!
Other AI improvements and a bit of Ranch gameplay
Enemy AI continues to be worked on by Chris, and although a lot of it is happening under-the-hood or is otherwise difficult to showcase, we're happy to report enemies react to grenades much more appropriately in V1035, as shown in this clip:[previewyoutube][/previewyoutube]
And here's a little co-op footage to show off a little more of new map Ranch and its audio ambiance, as well as the G36K in action:
[previewyoutube][/previewyoutube]
Map performance, lighting and color tweaks
Environmental artist Will has taken a break from the 747 remake to do an optimization and visual pass on maps.
Although subtle in some ways, this iteration greatly improves the look and atmosphere of the maps by carefully tweaking the lighting, color-grading materials and textures, and adjusting the fog intensity and color. Vistas (the distant parts of maps that are not accessible) were also upgraded here and there.
Let's take a look at some before (left) vs. after (right) shots:
As you can see, colors are more consistent (with fewer instances of bright colors standing out), and the fog adds a greater sense of depth — making it easier to judge different planes of the scene (near vs. far) and "softening" the more distant parts of the map which are naturally rendered in less detail.
💡 Fog color allows maps to have more unique atmospheres, especially with the sun at lower angles. Our lighting and weather systems allows different settings to be used at night, dawn, dusk and daytime. The fog color in the examples above is one of the factors for the "colder" tone during the day, but at dawn and dusk there is a very warm fog effect.
Will has also given Small Town a little facelift, improving the texturing and adding details and props. Here are some more comparison shots:
💡 Will continues to work on the new version of 747, as part of the big push to update all of GROUND BRANCH maps to the same visual standards as Depot, Small Town and others.
Another look at some of the new primary weapons
For your appreciation, media wizard Rangda has put together a few gifs of the LVAW, G36K and AUG A3 being fired and reloaded. That's it — just a little eye candy:
[hr][/hr]
This has been Intel Report #025!
And that's all we have for you today.
As mentioned earlier, there is more going on in V1035 than we're showing here. Some of the other stuff we've already shown you in past intel reports (be sure to check them out if you haven't already), and there are a few other things that we're choosing to not reveal before the actual release, because we feel it's important to keep a sense of mystery in our relationship.
Stay tuned, and we’ll see you again soon!
AI spawn points now have a "Day Spawn" and "Night Spawn" option, determining when the spawn is used
When spawn points are disabled due to day/night settings, they change color
The times when day and night transition occurs are set in the mission settings panel. They can be linked to astronomical time (standard hours), e.g. linked to sunset and sunrise
The AI loadout selection has now been streamlined for ease of use and visibility
More logging is provided to help identify issues with mission loadout setups
Idle tasks can now be set in the spawn point settings (work in progress)
New AN/PVS-31A night vision goggles model, seen here with the (also new) AUG A3
