BoredGamer

Star Citizen, Squadron 42 & Theatres of War News, Guides, Videos & Gameplay by BoredGamer

Star Citizen Makes Fantastic AI Progress While Squadron 42 Marches On…

Star Citizen’s AI is now making fantastic Progress While Squadron 42 Development Marches On… we find out what CI have been working on over the last few weeks in regards to their Single Player Campaign Sq42 and the updates that both SQ and Star Citizen have made with it’s AI and what the various development teams are focused on now in September.

AI Content focused on new medical behaviors, creating three new usables:

  • Sitting Angled Console – a desk where NPCs can sit and do admin work, such as checking on patient’s medical records
  • Examine – Spots on either side of the medical bed allowing doctors to approach a patient and diagnose their condition. This will include the use of both datapad and hand scanners
  • Medical Fridge – where medicine is stored. For now, NPCs will visually inspect the stock level and enter the information on the datapad. Eventually, they will restock it when quantities are low

They also continued to work on the chowline usables and resumed work on the food vendor from last year.

Alongside this, AI Content continued to improve general locomotion realism by taking the overlay animations worked on last month and categorizing them by frequency. For example, coughs and sneezes will play 10% of the time, head stretches 40%, and more extensive, bolder actions like checking the mobiGlas will happen 30% of the time. They’re currently designing a priority system to trigger these animations.

The AI Feature Team are implementing the traits system, which allows them to specialize AI characters using behavior logic by either limiting or favoring certain behaviors. For example, an AI with the ‘cautious’ trait will prefer moving from cover to cover when approaching the target, whereas with the ‘aggressive’ trait will directly move towards the target ignoring available cover. Other traits include ‘gunners’ who prefer using turrets, ‘medics’ who aid allies by healing them, and ‘addicts’ who use stim drugs whenever available. The core system has been implemented and several traits are now in place. Further traits will be implemented once the corresponding behaviors are developed.

With the basis of the ‘investigation’ behavior implemented, the team moved on to implementing group investigation, which involves sharing potential hiding locations between AI characters searching for the same target. This behavior is built from the token system that allows multiple AIs to collaborate by pooling resources and was developed with future use in mind. The generic shared data, in this case hiding location, can be specialized for specific use cases. As the NPCs move around the room, they determine and share which hiding locations have been seen so that the other NPCs won’t revisit them. From a simple implementation comes quite complicated results, with NPC’s now moving to cover the room as you would expect them to do in real life.

A new system related to both features has been implemented : the firing token. When several AIs are targeting the same character and want to fire, they request one of a limited number of shared tokens. If they’re successful, they will be able to fire. However, if there are no tokens available, they will not be able to fire and may consider alternative behaviors, such as moving to cover. This allows the team to further control the pressure placed on the player while also generating covering-fire behaviors, as some characters will fire whilst others are moving.

Traits were also implemented to suppress requiring the firing token in different ways. The first is to ignore the firing token completely, which means that the character will always be able to fire without reducing the number of available tokens, which can be used for boss characters. The second is to allow a specific AI character to always get a firing token, giving them firing priority without increasing the number of firing enemies.

Development of the AI perception system continued throughout August, ensuring that the escalation of threats in the perception meter can be controlled by tweaking the setup in Dataforge. 

A new sixth-sense perception range was also developed to control gameplay when the player is sneaking up behind enemy characters. The perception meter now allows the devs to generate gameplay for stealth kills, rather than stealth killing being an easy option, as the player must minimize the amount of time they’re in close range before they’re noticed.

On the animation side, AI Features polished female Human combat animations and created block-out animations for the improved sharp-turn functionality. They also continued polishing Vanduul search-and-investigation animations, including specific search locomotion animations.

They Said – “The AI Feature team have been hard at work developing lots of different features. This is partly due to the strong foundations of our AI code that were developed by spending a long time thinking about future functionality. As a team, we’ve also grown and have been improving the communication between our team and the designers through shared language, documentation, and regular meetings. Through this, we’ve able to take their vision and make it a reality” 

AI Tech continued to extend and implement new functionalities for NPCs traversing navigation areas. One improvement allows the designers to specify the opening width for a door navigation link and use that information during the pathfinding step. Now, NPCs will use the entire width of a door, not only the center.

They also began extending ladder functionality by creating new navigation link adapters and movement blocks so that AI characters can use them in similar ways to the players.

AI Tech also progressed with the NPC movement refactor. Specifically, they worked on a better separation of how logic is processed when animated character and actor states are updated. This allows them to correctly handle the actor LOD system, which can have both parts of a character updating at different rates. The overall movement refactor also went through a review process and intensive QA testing in preparation for its release.

Further iterations were made on the NPC seamless transitions prototype. This involved ironing out the small position and pose pops when handling animation control between different systems 

and adding support for selecting the best usable enter animation based on which has the best enter pose requiring the least warping.

The Subsumption editor tool was opened up to more designers so that it could go through more intensive and in-depth functionality and performance testing. Bugs and feedback from users were addressed and performance improvements began when loading bigger mission graphs.

A new feature for ship combat behaviors, ship-pilot perception, was extended to include vision alongside radar signals. This will allow AI ships to react to hostile characters on foot and engage them in combat.

AI Tech continued developing and improving reinforcement and disembarking behaviors. Part of this involved allowing missions to designate points where the squad group will go and investigate in case hostiles aren’t visible.

Animation worked on useable first reactions, ensuring all useables have a reaction set assigned to them. They also continued with the female-spec-ops work to get female buddy and enemy AIs in-game.

EVA and zero-g were further developed, with focus on attaching characters and moving through tight spaces. Work on helmet use continued, as did ladder improvements.

Vanduul searching for players, the chowline, medical bed, and custom animation locomotion work progressed throughout the month too. Tasks were completed for worker facial animations and scenes, including assets for the med table.

Part of the team continued to solve mo-cap from their backlog, the majority worked on the stage build and prepping for Narrative and Marketing shoots.

Character Art and Tech Art continued to develop the Screaming Galsons armor and the remaining Navy uniforms. They’re also working on an internal head-scan and cleanup process, proving out a new pipeline using tools developed by the Tech Animation team.

Tech Animation – began consolidating all the new heads they would like to appear in-game. This involves processing a lot of old and new data along with alignment to deliver on the requirements for both parts of the game. So far, 60 heads need to be scanned, created, rigged, and integrated into the DNA gene pool to achieve the vision.

Weapons Art continued to develop the Volt weapons mentioned in last month’s report, with the main modeling pass completed in August. They then moved on to refining the iron sights to improve the onscreen positioning.

New Screaming Galsons melee weapons were greyboxed and are currently having their setup and initial animations worked on to ensure the new metrics work well.

Several new skins were kicked off and the team also converted a number of weapons to the current tint system to allow for the creation of future skins.

VFX Concept Art completed the first pass on new weapon effects and helped the artists visualize a complex scene involving a powerful energy source. 

Gameplay Features working on a new framework for the mobiGlas, which will make better use of Building Blocks and the new UI cards the Tech team have been developing. This will then host all the new apps being built for SQ42.

New functionality was added to determine when players are in social areas and change the field of view to differentiate them from combat situations. For player hints and tutorials, new conditions were added to give the designers additional flexibility when they want to trigger them.

A new system is also being developed to allow the player to control overhead cranes. This could be expanded to allow players to remotely control moving drones too.

The FPS Level design team began taking five chapters to a more complete state with a continued focus on stability. They also tested the first third of the game, with the intention of giving players a seamless playthrough.

Like the FPS team, the Space/Dogfight team continued to focus on getting a large part of the game fully playable with functioning systems and mechanics. Alongside creating a seamless experience, they continued with scene set-up, bug fixing, and addressing review feedback, though the main focus was on refactoring and streamlining several interstitials to the required standards. This will make all tasks easier, from working on chapters to devising checkpoints.

Gameplay Story worked on two new scenes: The first being a piece that connects to a cinematic scene and features marines cheering. The second involves a medic helping a wounded miner.

Following feedback, they also created a new tram arrival with mo-cap being planned to allow two key characters to board it smoothly.

Narrative planned any additional content they’d need to capture.

Reviewed and reiterated on various sections of gameplay.

Cinematics completed the implementation work for major sequences in chapters one, five, and seven. This included lengthening the opening shot of the campaign. 

The Vanduul movement style was also finalized and prior block outs received more fluid motion and precise timing. Adjustments were made to a few scenes on the Javelin bridge, which was recently lengthened for FPS combat. 

Further work went into the timing of a large weapon too. 

For chapter five, further work was done on a tram ride, with the team working alongside the physics programmers to get litter, handles, and characters to sway when the tram accelerates. Other work involved a mining tick landing and walking to be unloaded, a meet up with a friendly character, an elevator ride, and a tram accident.

A scene in chapter seven received polish, which included a lot of smaller details. These included handcuffs, a Multi-Tool folding out and cutting, and the handing over of a med pen that required a small amount of new code from the SQ42 Feature team.

“Usually, when creating scenes with characters grabbing, unholstering, giving, or using items, we aim to solve it with proper systemic implementation. So, the actor action track in the sequencer tells the character to actually unholster the item, grab it, then hand it over to the other character. In this scene, we had the added difficulty of a stocked weapon being held while giving the med pen. 

In this instance, the user action confused the giver and given hand, so we often ended up handing over a rifle instead of a med pen. A small code fix to specify left and right hands for both giver and receiver fixed this.” Cinematics Team

The team also further worked on chapter 11, which needs multiple destruction set pieces. So, the VFX team provided sophisticated Houdini-authored simulated destruction ‘.CGAs’ that can house up to a thousand debris pieces and still run relatively lightweight vs a typical brute-force Alembic cache. They were then added alongside adjusted camera and timing shots to frame the scene.

Boom, that’s it for you Squadron 42 and Star Citizen AI Monthly Report. I am interested to know what you think? When do you think we could see Squadron 42 Episode 1? What do you think of those improvements made to AI… will they make their way to the PU next year? Whatever your thoughts I’d love to hear from you in the comments below.