I’m pleased to announce the release of our new home page: Boss Golf
This page will now strictly become the development blog for the game. Our home page will now contain more information regarding release dates, testing, rewards etc.
In anticipation for the release of the AI demo, I’ve also decided to improve the pipeline for testing the game. Moving to a closed alpha system so that I can keep better track of how things are, and get better feedback from the fans.
If you are interested in alpha testing Boss Golf, please check out the information on our closed alpha page!
So I’ve figured out how to improve the interface for the new terrain controllers for Boss Golf, and I added a button to display a grid overlay on top of the terrain, so that you can better gauge how it all looks!
Mainly, the painting tool now has an overlay on the terrain letting you know the area that will be affected. You can see it in action in the gifs below:
And the new grid overlay, which makes understanding the terrain slopes even easier:
That’s it for today’s mini-update! There’s a holiday coming up this week, which I plan to take advantage of to further the development of the AI in the game. Still hopeful to have little AI golfers walking around the course and playing a nice game of golf!
Oh yeah, and as always, you can find the updated demo Here!
I’ve just about finished connecting the new terrain system with the UI. There are some nags here and there, as is expected, and the interface for the vertex editing doesn’t work perfectly, but it’s all operational.
For this demo, however, I’ve removed the golf playing feature as it is due for a revamp in order to accommodate the AI system. As a result of that, I’ve also left the previous demo online in case you wanna try that part of Boss Golf.
And you’ll find that your play area has been largely expanded from before. And you’ll notice you won’t be able to build everywhere: that’s part of the course being only allowed certain parcels of land when you start out, with more being unlocked as you buy them and whatnot. The new chunking system will allow for that to be done pretty straightforwardly.
Without further ado, you can download the latest version HERE!
And here’s some things I cooked up with it:
That’s it for today, folks!
Will continue working smoothing out the implementation (and damning Unity for not supporting gizmos in runtime! (probably with good reason)), tweaking the shooting mechanism, and getting the AI in there!
I’ve finished coding the improved terrain system and the actions. Now all that’s left is connecting it to the UI!
But first, I’ll give you a preview of what’s coming. First, you have the good old drag to place the decoration:
Select the tree type, drag the tiles and the trees will be placed. Each tile can support up to 4 decoration pieces. Adding anything after that will result in the decorations simply being shuffled around.
If you want to fix it up, I’ve added a trimming tool that currently removes the last placed decoration object from the tile:
Later, I’ll make it so it actually aims at the decoration object you’re aiming at and removes that.
Then, for elevating the terrain, I’ve reworked the previous options to use the vertex directly. For altering the elevation directly, you can either click and drag (for fine-tuning), or simply click and hold (for painting). For each of them, you can choose between 1 vertexes, 4 vertexes (a tile), or 16 vertexes (the tile and the neighbours) as below:
And for painting:
And finally flattening:
As you may have noticed, the decorations nicely go up and down as the elevation is changed. This is part of the update to the decoration system, which makes it work much better than before.
So that’s what I’ve been working on this week! I’ll finish hooking it all up to the UI tomorrow, and perhaps release an updated version of the demo so you can try it out. And then go back to the AI implementation!
(Ah! Forgot to add: I’ll be turning off the water tile for now as it’ll need a deeper reworking. I’ll just replace it with a regular blue tile so that you can still plan on how the water would feel.)
Quick update on the implementation of the new terrain for Boss Golf!
Look at this:
That, my friends, is a massive terrain composed of a 20×20 grid of chunks, each chunk containing 16×16 tiles. That’s roughly 102400 tiles. All much more easily manageable, and with better performance than before. Now I can actually make the terrain surrounding the golf courses so that you no longer can peek at the end of the world!
As explained before, this new system relies more on preprocessing the grid data, so that alterations to vertices and UVs are much more quickly resolved than before. No more loops inside loops inside loops!
Just finished implementing the UV mapping to it, as well as the mapping of surrounding tiles and vertices. No seams when you move vertices between chunks. Already have some plans for new terrain tools too, to give you more possibilities when editing your golf resort course!
Next up is connecting it to the existing controls, and changing how the decoration system works so that placement is handled better and performance is improved.
As promised, today I’m going to talk about the Memory component of our AI golfers!
So far, we’ve taken a look at how the behavior is controlled via State Machines; we’ve looked at how we’ll use Utility to choose the best course of action; and we’ve also taken a look at how the AI will execute Actions through a system of intentions.
But we want our AI to be smart; this means that we want the AI to be able to learn from its past mistakes in order to improve its score and rating. And we do that through memory!
Memory will be separated by holes. The AI will record the intention of the shots it took, the location of each shot, rate the outcome, and the overall performance for any given hole. Next time the AI plays the same hole, it will generate the actions as usual, but it will also compare with the previous taken in similar circumstances, and compare the utility value of that action against the outcome, and weigh it against current circumstances.
A simple example to illustrate the system is driving off the hole below:
In the image above, we have three shooting options that the AI came up with.
The top orange option is the shot that the AI took the last time it went around the course
The bottom orange option is the shot that the AI calculated as the best for this time around
The purple shot is what actually happened
So in the situation above, we have an AI that remembered that the last time around this hole, it took an awesome shot through the left side of the fairway, stopping quite near the edge of the water but within awesome range of the green.
For the other options that the AI came up with, among them the bottom orange one, through the right side of the fairway, was the one that the AI felt most promising at this time.
The outcome of the left-side action was stronger than the predicted outcome for the right-side action. The AI evaluates this. It looks into its memory for the previous outcome, and compares with what it expects of the other outcomes. In this case, it is pretty clear that it is the superior action.
But, alas, just like us humans, the AI will also remember the past in a different light.
The last time it played that particular hole, it had a super drive right of the tee, exceeding its expectations and being regarded as a “star memory”, meaning it values it far higher than what it should. So memories aren’t just a 1 to 1 remembrance of the past: they also have their fair share of rose-tinted glasses. If an AI does better than it expects on a shot, its memory will also be colored by that event. Granted, this will depend on mental attributes for the AI, but it will be a likely occurrence.
And more than that, the accuracy of the memory also decays with time. Since the last time the AI played this hole was months before the current moment, the outcome will be slightly mistaken, as will the parameters the AI used to take the shot. In this case for this particular AI character, he will overestimate his strength, he will mistaken his position, and he will end up landing the ball square on the bunker trap. This will now take the place as his most recent memory of the driving stroke of this hole. He won’t immediately forget his previous memory; it’ll just have a weaker connection since the next attempt at that resulted in catastrophe.
Next time he plays around, he’ll likely try something new, or he’ll try the previous previous memory (IE: the top orange stroke) but will be more careful this time around.
The memory system will also work to help with the knowledge representation of the hole. The AI will remember each hole in a tile/type/landscape system, so that it can always compare the hole to the memory it has of the hole. So if the hole has changed, either through your own landscaping efforts or otherwise, the AI will know, will disregard the memories of past strokes a bit more, and will focus on trying something new.
This should end with an AI that is always trying to familiarize itself with your course, and has past performance to base future performances off of. It’ll also help in making the AI decide its feelings towards a particular hole (phrasing!), so that it can react to the changes you make to your course.
So I hope you enjoyed this overview! It’s very high level, but it should give you some clarity as to how the AI will work. Now we have the 4 main systems sketched out, I can continue implementing them. Will be done with the new terrain this weekend, then I can focus on finishing the first pass of the AI. For this month, the goal is to have AI golfers playing with your holes (phrasing!) so that I can test out some basics of the systems.
As I’ve said previously, while coming up with the concept and the systems for AI, I’ve also decided to rework a bit of the terrain features in order to make it smarter and improve performance.
So I’m almost done plugging in the new terrain system into the game. This time around, I went a bit smarter with my separation of concerns. The current system in the game features a breakdown of the mesh into tiles, quads, triangles, vertices, all controlled by a single TerrainController script which manages the actions on each tile.
It works ok, but it’s a bit overcomplicated in certain areas, and too simple in others. The point that really sticks out is when it has to update the mesh after any alteration is done to it.
Basically, it is currently rebuilding the vertice/triangle/UV list every time you do an alteration. This isn’t ideal, as it contains loops inside loops querying data that hasn’t been changed to find the ones that have been. Plus the current system is also less conducive to having multiple chunks and altering between them. It would be fine if we just stuck to one chunk and made it massive, but that’s not ideal specially if we want really massive levels and easier to manage data.
So, enter the new system! This time around, I have a SmartMesh class who holds the reference to the Mesh component itself, as well as a TerrainChunk component that holds it all together, managed by a TerrainBase class that can decide how many chunks will be spawned and the dimensions for each chunk. Reworking the input management as well means cleaner code and easier to expand. Most important is that all vertex data is held by TerrainVertice objects, which can be directly accessed either y the TerrainChunk or the SmartMesh. The SmartMesh also generates a mesh on startup of the position/index of each TerrainVertice. So what dos this mean?
It means that when a TerrainVertice is changed, it changes its position, and it tells the SmartMesh which index of the vertex list needs to be updated. This is done, and the new vertex list is assigned to the Mesh component. Hey presto; no more nasty loops within loops within loops!
Similar concept will be applied to UV and other aspects of the terrain. So the end result is a cleaner system that’s more manageable. I’ve also picked up more knowledge as I’ve progressed with development, and it has been ideal to implement it now.
Oh, and you’ll also be able to “sculpt” the terrain better, and I’m also reworking the height limitations of the course so you can have nicer features. Check out the simple demo below:
You’ll be able to “paint” the heights with the new tool. (Also keeping the old options too for those who prefer them). And since the vertices are easier to manage, I’m also updating the height limitations to give you more freedom to build taller things, and smoothing out the process: if you move a vertice too far above a threshold, the neighbouring vertices will also move up, creating a more organic flow.
I should be done with this work by the weekend. At the same time, I’m also implementing the AI system, so stay tuned for that too!
Tomorrow I’ll make a post about the memory system!
So far, we’ve described the state machine used for keeping track and modifying the behavior of the AI, and the utility function that is used to select what course of action the AI should take.
And now it’s time to talk about the other piece of the AI puzzle: Actions!
Put simply, these are the functions that the AI executes to perform an action in game; things like moving, idling, shooting etc. All of those actions are self contained in their own methods, take a parameter or three, and execute until the action is successful or fails.
Of special interest to Boss Golf is the shooting action. In golf, there are countless ways that you can hit your golf ball. You can add some backspin, you can give it some fade, you can try to go straight, you can go for a higher angle, you can go full force etc. Your success with that action, however, depends on myriad factors like your skill level, your confidence, your performance, the weather…
Therefore, for Boss Golf, when it comes to the shooting action, I’ll implement another subsystem that is based on intention, that is, the AI will tell the gameplay system what sort of shooting action it plans on executing, and the gameplay system will take all of the aforementioned factors and calculate the outcome for the AI. This will give more “human” results to the actions of the AI, as they would never be able to hit the same shot exactly the same way twice. There will always be deviations.
Breaking the system down:
The AI would come up with a handful of potential shooting actions for a given situation; it would think of a route by hammering it over the pond, another route by cutting through the trees, and another route by just hitting the fairway.
It would then parse those actions through the utility system, to see what action would be the one with the best potential reward/utility for the AI
After it found the best course of action, it would communicate the intention to the gameplay system, telling it it wants to:
Hit ball with club X
Hit ball with strength Y
Aim shot at square Z
Add X amount of backspin
The gameplay system would then modify these attributes based on the environmental factors surrounding the AI
If the AI is lacking confidence, add some uncertainty to the aiming
If the AI is behind in the championship, add some extra strength due to nerves
If the AI doesn’t have great skill, add way more backspin than the AI intended
If the AI has a lucky modifier, alter the shot slightly so that it lands in a slightly better position
The action is executed, the AI hits the ball with the modified intention, and analyzes the result
Of course, this is a higher level breakdown of what is going on behind the scenes, to better illustrate to you the capability of the AI. There will be a myriad other factors that can potential add a modifier to the AI’s decision. Our friend Jumper made a very nice write-up on our discord channel, giving his take as a sports psychologist on the mind of the players and the many factors that can affect their performance on the pitch.
Now we know how to manage behavior, how to make decisions, and how to act with intention. There’s one more piece of the puzzle missing: Memory! As golfers go around the course, we want them to remember the way they tackled a hole before, so that they can try and improve on it or follow their old steps. I’ll break this feature down on my next post!
That’s it for today’s update, folks!
As an addendum, I’ve been hard at work implementing the AI, and also redoing part of the terrain system as it’s grown to be too bloated and not smart enough for the AI to work with. I’ll be done with that by the next weekend, and hopefully have some AI gameplay to show as well. Such is the nature of game development: the best ideas always come after you’ve implemented a system. This new structure will make it easy for the AI to evaluate the shots, as well as to place decorations and other structures on the course.
Today I’ll be talking about another part of the AI of Boss Golf: Utility
Simply put, Utility is a value that determines how useful an action is for the AI to take at a specific moment in time. For example:
Take an enemy AI that has a health value and an ammo value. And it has only three possible actions:
At the start of a conflict with a player, if the AI is at full health and full ammo, then the utility value of the first two action is pretty much zero: Shooting would be a much more useful action in that specific moment.
As the AI engages in combat with player, things get more murky. If the ammo is approaching zero, then the utility of the reload action increases; the act of reloading becomes more important than shooting as you need bullets in order to shoot. At the same time, as the health of the AI decreases, the utility factor of the use healthpack action increases, as the AI needs health otherwise it will die.
The way this calculation is done depends a lot on the factors and data that the AI has access to. If it knows that the player is at low health, then reloading first instead of healing might be a better option. If it’s a particularly coward enemy, it will give priority to heal itself instead.
And this is just with three possible actions; add in more actions, such as running, finding cover, dodging etc, and you get much more variety of behavior.
The same thing will happen in Boss Golf, specifically when it comes to deciding the type of shot to take along a hole. Let’s use a sample hole to illustrate how it would work:
In the Par 3 above, hitting the green from the start seems like a solid choice. The hole is well within the 150 yard range, which is achievable by most golfers that take to the course. But even pros make mistake; and there’s a couple of factors that can affect the success of a stroke:
Performance so far
Wind can drastically affect a stroke, making your ball go lower than expected, slower than expected, veer laterally.
Confidence can affect the quality of your stroke. Lower confidence can lead to slice/hook, lack of power, reduce your accuracy, among other things.
Performance so far. If the golfer has been having a great day, he’ll be more likely to take risks; If he’s playing his ultimate best, he might be more risk-averse in order to keep his good handicap; if he’s having a bad day, he’ll take even less chances in order to try and recover.
All of those attributes (and more) would affect the calculation for utility of the shot. For Boss Golf, the golfers would always calculate a specific number of shot (based on their attributes), calculate the utility for them, and make the choice.
Below I’ve simulated some potential shots and their results, to give you an idea of what could happen:
Some golfers would try and keep to the straight brown shot, getting very close to the hole. Others might go the blue route since they’re lacking the confidence they’d be able to hit the ball across the lake in one go. Some might overshoot it into the green following the pink line because they believe they needed more power. The guy in the green line may have messed up his first shot due to lack of confidence. And the poor fella on the white line just sent it straight into the water hazard.
After taking the shot, the result would also affect his current attributes, affecting his next shooting decisions as a result. So if the shot landed where the golfer wanted or better, his confidence will increase, so he’ll be more prone to taking a risky shot next; by the same token, if the shot ended up going badly, his confidence could decrease, leading him to be more risk averse on his next shot.
And that’s about it. Now we know how the AI will be controlled (via State Machines) and we know how he’ll decide which action to take (via Utility).
But we’re still missing one key piece of the puzzle: the Actions themselves.
Boss Golf features an action system that limits the possibility/quality of actions by the attributes/ability of the golfer. A pro-level AI golfer will have more tools (read: shots) at his disposal than a first-timer out in the tee. He’ll know when to pull, when to push, when to chip, when to approach, and be able to use these actions when out in the green, whereas a first-timer will be limited to less imaginative and simpler shots. As they grow in skill, they’ll be able to add more shots to their arsenal.
All of that will have to be factored in by the player when designing the course. If it’s catered mainly to beginners, it makes little sense to have very difficult courses; if it’s catered to professionals, it would make little sense to have too many easy holes. This balance will be vital in the development of your golf course.
I’ll make a write-up about the actions on my next post this week. Writing about the systems like this has been very helpful in aiding me with visualizing how the system works and building it!
So expect AI integration by the end of this month, bringing life to your course!
To break the silence today, I’ll talk about the sketching of the AI of the golfers in Boss Golf, something that I’ve been working on over the past week!
I have a great interest in AI, being an AI programmer professionally, so for Boss Golf I want to create some very interesting things with AI, to reinforce the living feeling of the NPCs, as well as give rise to the unpredictable events in the game.
To start off, I’ll describe the main backbone of the AI in Boss Golf: State Machines.
State Machines are basically a set of nodes that determine the current state of the AI (sleeping, idling etc), and how they transition into each other based on set instructions. How/when these transitions happen are decided by the programmer.
For Boss Golf, since gameplay is very segmented, state machines will be used to determine the basic state that the AI is in. Some of these states include, but are not limited to; teeing, putting, approaching, idling, waiting, interacting, resting, moving. The more expanded the state machine is, the more different and interesting behaviors we can get out of the AI.
For example: why separate the act of playing golf into three separate states? (Teeing, Approaching, Puttting) They could easily be condensed into one “playing” state since the actions that the AI will perform are very similar. (moving and hitting the ball)
The answer is that by adding this degree of separation to the behavior, we can have the AI interact and respond differently to events that happen at different points of play, in a more neat or organized way. If rain interrupts play during teeing off, it’ll have less impact on a golfer’s mentality than if he was putting; the stakes are different. Instead of having to keep track of how many strokes he has and how close he is to the hole, we can simply query the current state, and make the AI react accordingly.
And that’s just on one particular possible event. When you combine with other events, and add layers of actions and behaviors, you end up with a more interesting AI rather than one that simple goes around hitting the ball.
We can even add some variety to the transitions based on randomness or the attributes of the AI character. A golfer with poor health/lower stamina may have a bigger chance of switching to the rest state after a hole than a golfer in better health; a golfer with more aggressiveness maybhave a bigger chance of switching to a raging state after a failed easy shot than a calmer golfer.
And that’s the basics of it. I’ve sketched out some basic states that need to be in the game, and am in the process of implementing them. With state machines, we can also add new states at any point in time since they are self contained, so expanding the AI can be done quite easily.
Next time I’ll talk about another aspect of the AI that requires more difficult work to get it working correctly: Utility Functions.