[p]Hey Exanimates,


We quickly recorded a video at the last minute to showcase some of the improvements discussed in this diary, but read on for a more in depth and accurate explanation.

[dynamiclink href="https://www.youtube.com/watch?v=OheXErPzAeA"][/dynamiclink]
We last talked about some difficulties we encountered with physics for a new creature, and how despite our efforts to get it working with some incremental improvements and adjustments, it became clear that some aspects of our character physics required a significant overhaul. While it's unfortunate that it's taken so much additional work (no one wants the work on this update to stop more than us!), this has led to some really important improvements to the game overall and how it plays. Considering how difficult this character physics work is and how much time we've spent on these things in the past with varying levels of success, we got through these many and big improvements surprisingly quickly.

After we'd managed to at least cover some basics like good muscle function and control and moving on flat ground without constantly face planting, the next goal was traversing uneven ground, slopes, stairs etc. This was already problematic even for human characters, there's a lot tripping or stumbling over even minor obstacles or bumps in the ground. We know this is something that players can find quite annoying and it produces a lot of clumsy and silly moments in general. This has only become more of an issue as we've introduced more verticality and complex environments. We've made some attempts to improve this over the years, but the improvements have been marginal. This new creature really couldn't handle the clumsy movement and it was time to give properly solving these issues our best shot.

We've failed so far because it is indeed a very difficult problem. Our physics means that the feet collide with every little thing, if the character fails to avoid hitting something with their foot, they will trip and lose balance. The "ground" isn't some simplified navigation mesh, it's really a bunch of random polygons, it could be a pile of dynamic objects. The character isn't just some 3D animated sprite gliding over it, it needs to use muscles and generate forces well ahead of time to eventually clear obstacles, ascend and descend slopes, stairs etc., not just with their legs, but the entire body. This requires scanning the ground ahead, trying to determine its shape, predicting what adjustments will be necessary to the animation to traverse it, how muscles should contract to produce those motions, how hard to push off the ground, how fast it can all be realistically done and so on. Besides the mechanical problems, it also has to look natural. If the motions are exaggerated or too fast, if the character does things like lift their feet too high to clear small bumps, it looks silly or just very weird. It all has to be measured just right and very smooth. And that's just the basics, movement must work on any type of slanted or uneven ground with perhaps dips and holes, there's no artificial barriers so anything goes and characters must react to tripping, stumbling and slipping and try to recover their balance.

Hopefully that gives you an idea of the scope of the problem. It's such a complex problem with so many moving parts with variable outcomes over time, there isn't some exact solution that you can realistically just work out. We started by laying down some foundations for better scanning and mapping the shape of the environment ahead, which was already a difficult problem, more so because of performance constraints with potentially lots of characters running around at once (whether you see them or not, physics don't disappear). We also made a new system for controlling muscles that would be better able to make precise adjustments to pre-designed motions like running. We also finally tackled an old problem with how strongly characters push off the ground to counter gravity or generate vertical motion to try to stay at the right height. A good start, but these are just building blocks and not a solution.

The next step was adding some visual aids for how we were scanning the ground, what the prediction targets were, when and where exactly the character decided to plant each foot and so on. From here we were able to fine tune prediction mostly through trial and error, diagnose problems and make various improvements, making dynamic adjustments to mechanical states based on what was physically happening rather than what was expected in the reference animation. Eventually we started to see some good results. Characters were negotiating difficult ground and smartly stepping or hopping over most obstacles, our new monster was climbing over stuff like a champ and it all looked good and smooth. This can all be made even better, but fine tuning these things takes time and we're in a hurry, so it was time to move on. With everything working so differently there were a lot of cascading effects and other things that didn't work anymore or we had to outright disable just to test this.

Our next problem was the dynamic balancer, how we detect that the character is losing balance, tripping or slipping so they can try to do something about it, or understand when they're not able to negotiate particularly difficult ground so they can adjust or give up gracefully. There's a lot of potential goofiness here, but a big problem was that the dynamic balancer must smoothly take over, sometimes briefly and barely assisting, so these must interact seamlessly, but walking and running were now using a completely new system for dynamic behaviours from the balancer which is also the foundation of all combat and other movement. The logical thing to do was to make everything work on the new improved systems rather than try to maintain both and make them communicate, but this meant changing combat footwork again, which as we've noted before, is scary ground as it's such a fine tuned system and critical to the feel of combat.

Eventually we got the footwork seemingly working well using the new systems, we thought we had almost everything working and moved on to testing the new creature's combat AI. But in testing combat we noticed the footwork wasn't there, it felt a bit sluggish and unresponsive, looked clumsy at times and characters were falling over too easily. This was discouraging, it was quite possible that while some things were technically better, the flaws of the old system are what gave the footwork its "magic". Eventually with a lot of A-B testing between the new and old footwork we were able to identify the differences and return the footwork to its former glory, with a few notable improvements.

Another thing that broke from all the changes was crouching. Crouching is one those "WIP" features, and it already wasn't working well. A proper crouching implementation would need more work, but we were able to do something that is still not perfect, but definitely better than it was. Characters can now crouch well on uneven ground, they can perform attacks better and even move around reasonably well. Some quite ugly problems were fixed.

We also had a separate problem with getting up from prone positions. You may have noticed that when getting up characters will occasionally slide or spin in weird ways. Of course this problem was much more pronounced with the new creature, which basically just started turning in place without ever managing to get up. This was again due to some artificial forces we used to try to get characters in the right orientation to perform their getting up move. It's another one of those problems we've had difficulties with in the past, but as we keep discovering with so many core improvements to our character physics these heavy handed hacks are not really needed anymore. We managed to work around the problem without the ugly hacks, so they're gone too. In future we plan to add more quick recoveries from different positions and partial falls, but not today.

We're trying not to bore you with too many details, but these are the broad strokes. It took a lot of failed experiments, dead ends, promising things abandoned for now due to time constraints and maddening persistent problems like ankles misbehaving or characters leaning too much and falling and so on. It's been a truly hectic month. We know we can improve a lot on these new foundations, but we've been trying to get through this as quickly as possible and moving on as soon as we could call it "good enough for now". Even so, these are huge improvements already, and a big hit on the backlog of difficult issues to solve. Considering all we got through it fast. Moving around feels so much smoother and there's a lot less weird, silly or irritating moments in general.

One last thing, we mentioned testing combat AI for the new creature, and this is quite an important topic. We want this new monster to be quite accurate with some of its attacks, and because it's also big we wanted it to be more deliberate in its actions and less like the typically quite hectic AI. As we explored ideas for this we realised that probably a lot of things, even humans, could benefit from some of this, and this led to some new designs for AI tactical decisions. We've been testing these ideas and we've now got a barebones, but fully functional version of this new combat AI which already seems to be a notable improvement. The combat is very dynamic and landing hits on an unpredictable human player is difficult, the AI tends to be quite spammy, if it had to wait for the player to make a mistake and give a good opening it wouldn't be much fun. AI behaviour is an absolutely crucial part of what makes the combat fun and engaging. This would be a massive topic of its own, but to keep things brief we've come up with a method to allow both hasty and impulsive actions and more precise and careful ones and to support more methodical behaviour without it feeling slow and dull. The new creature no longer felt too twitchy and hyperactive for its size, but still very threatening. It's still early, but we should be able to control how "energetic" fighters are, give them more varied and personalised styles of combat, add dynamic changes, make the AI feel more diverse and engaging and hopefully raise its skill ceiling. We expect this to really elevate the gameplay with more work down the line.

There's always many things that happen in a month of development that we can't realistically cover in these diaries, but this was a particularly productive month. There were setbacks, but with good outcomes. We're constantly entering unknown territory as we expand the game and we just don't know when we'll bump into an unexpected problem or how big it will be. Honestly sometimes they seem practically impossible, and sometimes our efforts come to nothing. So many things that might have very simple tried and tested solutions in many games can be immeasurably more complex in this one, and we're a tiny team developing everything from scratch where much larger projects use ready-made engines and solutions. We're fully devoted to our community's desire for something special and your countless hopes and dreams, nothing else matters to us.


Best,

Bare Mettle[/p]