Reverse the ‘Verse goes across the globe again to Foundry 42 Frankfurt, check it out.
TL;DR (Too Long; Didn't Read)
Today’s RTV was in Frankfurt hosted by Brian Chambers and two guests: Jason Cole, Lead Cinematic Animator, and Francesco Roccucci, Lead AI Programmer
They talked about Miles Ekhart who was the Quest giver in the 3.0 demo
Miles took about three weeks from shoot to implementation and this was the first time they’d done a cinematic quest giver like miles using the new tools and bug fixing those tools along the way.
It may take only two weeks per quest giver like miles and that was a generous time scale they said.
Miles won’t be the only cinematic quest giver for 3.0 as they’re scheduling another shoot specifically for 3.0.
Mile’s eye movement during interactions was the actor himself trying to be sketchy, the IK Look system has only the on/off feature and they’re working on having it blend naturally so the AI’s focus is on you instead of in your general direction.
Francesco Roccucci talked in depth about the Subsumption system, I highly recommend you watch it as there’s too much detail for me to summarise properly
Subsumption is a collection of systems that include behaviour of characters, missions and is what connects it to the games code.
The goal for subsumption is to make AI as life like as possible and what a player can do or go, so can an AI. Ship repairs, manning a console, etc.
The amount of AI in an area depends on the area itself, small enclosed spaces wont be packed, but larger spaces will have more AI.
The intent is to LOD AI to allow a large amount on screen at any given time, the physical model is what limits the amount, not the AI actions.
Currently they can have up to 40 AI on screen at any given time without optimisation.
And we’re live with RtV, Brian Chambers, Development Director for Frankfurt, is on.
First guest with Brian today is Jason Cole, Lead Cinematic Animator. He’ll discuss the quest giver a bit, and answer questions.
Miles Ekhart is the first quest-giver in-game. He was cool to get in, was good to see him in there. Within animation there are lots of dependencies; environments, lighting, character rigging, etc… but now they’re starting to get the tools in, integrating more Maya stuff because in Maya they can sync heads and rigs, and this was the first time they tested it live.
The tools are still being iterated on, but the base is working, so they can start seeing what it’s all like.
[Who played Miles?] Not sure the actor’s name. They can find his name and put it int the notes after. He did a few characters in the game.
[Do we know if all quest givers will be voiced?] That’s more of a design question. For 3.0 they’re planning on adding more dialogue and quest giver stuff. The more that gets synced, the more that’ll get added. There’s a lot. The amount of facial animation that’s attached to the things is fairly in-depth.
The more the tools become streamlined though, the easier they can add things in.
[Is Miles Ekhart the only cinematic quest giver in 3.0?] Probably not. They’re scheduling another shoot specifically to add stuff to 3.0, how much they’ll add they don’t know, but Chris intends to add more.
As they go through the process, sometimes attention has to shift though. The assumption right now is you’d have more than just Miles in 3.0, but more details on that’ll come in the next few weeks probably. That was the first time they’ve run their system for PU stuff, so they’ll be handing it off to the PU folks to do stuff with it.
Frankfurt will work in conjunction with animators in Austin and UK, train them up on the new tools etc…
[How is working with the team over there going? The team in Germany is fairly new] It’s good. The team in Frankfurt is really strong, UK is really good, and Austin is as well. They actually had more stuff to go into the PU. There was subsumption, people walking around in the space, walking up to the bar, drinking, and walking away, etc… that didn’t make it into the final build. There’s a good amount of stuff they’re doing. What they showed at the live event isn’t everything they had, it’s what they thought was representative of what looked good at the time.
There’s a lot more from studios that’s just about there, but they want it to be fully there before they present it.
[How long did it take from the moment you got the actor until you were done with the final version of the journal video, and how much improvement do you see possible on that pipeline?] They talked about the process a bit in the video, but duration of time, it didn’t take very long. They shot the quest giver stuff at the end of the last shoot, so they had 3 weeks to put everything together for Ekhart and the hawkers. That was within 3 weeks, and they were testing the system on top of that. It was all getting solved in-house. It was the first time they’ve used the tools. They ran into lots of issues in those few weeks. They’re updating the facial tools, so they can have more control over look IK and other things, but for Ekhart they had to go back to an older system because the NPC heads weren’t on the new system.
Hard to gauge the actual time, but it would take maybe two weeks at most, start to finish. That’s being very generous as well, and normally in those estimates, they have to put a little more time into the estimates just to foresee if there are any issues that come up. If they say two week, and they go ‘oh I think it’ll take us 7 days’, other tasks can come in that take priority, etc…
There’re some things that should take 2 weeks, but they only take 2 days, and vice versa.
[How many chicken burgers can Brian Chambers eat?] Probably more than anyone.
[Will quest content each time be procedural and random generated, or will each Miles give the same quest?] Not sure, that’s more design, but you’d think they’d stack the quests on Ekhart, for Jason and the animation team, it’s more ‘here’s the stuff to animate, here’s what you need to put together’. As far as how different and random the quests will be, that’s more a design question.
[Miles seemed to have evasive eye movement, was that acting or tech?] That’s acting. It was him trying to be sketchy. They didn’t’ have enough time to make it look better. Right now they can only turn Look IK on and off, eventually they’ll be able to trigger it. So he’ll be looking at you, but not making eye contact, etc… And it’s not only the eyes, it’s a portion of the upper body as well
What we saw was still a first pass, it will still get better. If you look at the stuff they put out 6-8 months ago, vs. now, it’s night and day. What they put out within 3 weeks, they didn’t spend any time polishing it. They could add a lot more, with a bit more time.
And that’s all for Jason. Short break, another AtV video will play, and they’ll be back.
And they’re back. With Brian now is Francesco Roccucci, Lead AI Programmer. He’ll talk about the navmesh and AI in general.
The navmesh is one of the more technical bits of information they’ve given out. With the navmesh work that was shown off, and knowing about the subsumption roadmap… Nav Mesh is a really important, really critical part, of Subsumption. at the end of the day movement is a very well solved problem with AI, but with different things moving in relation to each other, and with having AI navigate between different meshes and from different parts of the Universe is much more difficult, and isn’t something they’ve been able to achieve in a short amount of time.
They wish there’d been more time to show it in Gamescom, but people’ll see it in the next update.
[Quick explanation of what subsumption means?] Subsumption is a collection of systems. We call it what it is, it includes behaviour of characters, missions, it’s the systems that connect AI to the game code. It’s got navmesh, everything.
Activities are connected to jobs;. if you’re a cleaning guy in the ship, your activity is cleaning. Cleaning the floor, the kitchen, the Mess hall, etc…
The character will iterate between activities. Then there’s the mission system. To do a mission you need to have specific characters doing specific things. Then there’s a scheduler, saying it’s morning, you should work, but someone else is sleeping because they worked at night, and there’s a priority system, as well.
I missed a lot here, LOTS of detail
[navmesh was also on the exterior of the ship, will NPC’s be able to climb on the exterior of ships?] Yes. That’s the idea. It’s important that the AI can move anywhere the player can. Ships have some exits that allow players to go on top, so NPC’s need to be able to go there as well. If there’s a mission that has you go up and repair, or jump off, the NPC needs to be able to do that.
[Will AI on landing pads evade landing spacecraft?] AI will be able to land only in assigned pads that are free. If they can’t do it anymore, they’ll be notified. Same communication system as the players. Key is always on behaviour. If the brain of the AI is correct, it’ll do the same as a player.
[How many NPC’s are planned to be inside a station walking around?] Goal is to have as many as they can, and as many that make sense. Doesn’t make sense fitting 200 into a small area, but there are some large areas where you can have lots. Idea is to use a sort of LOD system for AI. AI itself is very cheap; subsumption is based on templates, and each AI is doing certain templates at certain scheduled times. AI isn’t the intensive part, it’s more the physical AI model, etc…
[Have you run any tests that about filling areas?] They tried to be smart, what they tried to do with the Gamescom demo, they were looking for 30-40 AI’s as the limit for right now, with no optimisations etc…
[How do nav,eshes work when it comes to 0g?] 0g is not a Nav mesh at that point. Nav Mesh represents a surface. If you wanted to use mag boots, it’d use a Nav mesh. In 0g the plan is to have an hierarchical path-finder. If you want to fly to another station, first step is getting to a ship, then flying, land, get out, and walk. If there’s 0g in the middle, they’d navigate the environment as 0g.
The big challenge is figuring out how to do things like enter separate ships while flying, etc… figuring out transitions and zoning with the AI is what they’ll have to deal with in the future.
[Can AI interact with more than one player at the same time?] That’s based on behaviour. The AI sees everything around, but what the behaviour will do will decide what the AI does. If you have two characters that want to talk, the behaviour will interact with one, then another. There are different query mechanics that AI can use as well, ‘give me the closest enemy’, ‘closest friend’ etc… but the behaviour of the AI determines who to deal with first.
[How will AI react to sudden changes in the environment? Decompression etc…?] Always based on behaviour. If you’re trying to do something in an environment, and the environment gets depressurised, they can create specific reactions to do that, it’d become the highest priority. Panic could happen, etc… gets to the ‘fight or flight’, base root of who each character is. Civilian characters won’t react the same as marines, etc… similar to how they’ve done if a friendly character dies next to you. if you’re a normal character, you freak out, cause you’re not expecting that, but if you’re in a combat situation and that happens, you won’t freak out. There’ll be critical changes and non-critical changes etc…
Looking at the amount of careers, and knowing that the careers that characters can pic, AI can pick those as well. Some people may have an aggressive behaviour, and may be in an aggressive career, etc… and a lot will depend on animations and voices, and they’re going to constantly add content. Not only animations and voices, but they’ll add more behaviours as well as they go too.
AI is constant, it’s something important that will help the game be alive. The AI is kind of the game, because the way you play with the game and the way the game plays with you is all about AI.
That’s Frankfurt wrapping up now. The Terrapin is currently still on sale, part 2 of the Terrapin Q&A will go up later today / tomorrow. Thanks go to everyone who went out to Gamescom, the support and the love and attention CIG are getting is great, it’s an awesome feeling being able to share the current progress. Thanks go to everyone for watching the first RtV out of Frankfurt, so thanks to Subscribers for allowing this content to be done.
David here, I highly recommend watching this RtV. There was a lot that went by too quickly to transcribe, and a lot of detail that’ll be missed by just reading the transcript. When you have time, watch the video. Cheers!
A polite Canadian who takes pride in making other peoples day brighter. He enjoys waffles with Maplesyrup, making delicious puns and striving for perfection in screaming at the T.V. during hockey games.
Erris is Canadian. He does some random things for Relay, no-one really knows what, but still they're stuck with him. He’s also written one Young Adult novel that he can’t stand, which can be found here.