Introducing Lillie is the Keeper

You’ve always wanted a lighthouse. A contained, liminal space at the edge of the world, between sea and sky. A literal beacon in the night. I can put one in your pocket.

Lillie is the Keeper (LitK) is an upcoming Myst-like game played in AR on an iPhone or iPad. You’ll take on the role of Lillie Pine, a 15 year old girl shipwrecked in an empty lighthouse following a mysterious calamity in 1905. With your iOS device, you’ll explore the lighthouse by moving about your real-world space.

The game is structured as a short story, beginning before it begins and ending after it ends, in eight play vignettes or levels. The main character, Lillie, suffers from intrusive thoughts, which we hear as an older, more severe, highly critical version of her own voice.

The game is built in Unity 2021.2 and runs in iOS 14, using AR Foundation 4.2 as a wrapper over AR Kit 5. It is a solo effort by the author, including writing, voice direction, modeling, coding, rigging, sound editing and visual design, and stars actors Sheri Lee (Hearts of New England), Robert Harrison (The Equalizer) and Melissa McCue-McGrath (BewilderBeasts).

Tiny Convoy: Systems Programming

Since I was mostly the programmer, I’d like to look at some of the systems that make the game engine work. But first…

Inheritance?

For me, what made this game possible was an email exchange with Michael Schmidt at Unity, in which he cleared up something critical. Since Unity regards every script as a different C# type, how can you subclass a script into different variations? More specifically, how can one script interact with another when it doesn’t know if it’s the class, one of the subclasses, or one of the subclass’s subclasses?

Crunchy McCrunch-Crunch

The key is that Unity will treat a subclass script as if it were any class up its chain of inheritance. When I subclass ActualThing into Upgrade, and Upgrade into Sensor, other scripts can reference a Sensor script as if it were an Upgrade script or an ActualThing script. So these lines are equivalent:

float currentMass = someGameObject.GetComponent<ActualThing>().mass;
float currentMass = someGameObject.GetComponent<Upgrade>().mass;
float currentMass = someGameObject.GetComponent<Sensor>().mass;

A Sensor script can then override, for instance, ActualThing’s takeDamage() function, and respond to it differently than, say, a rock would:

//in ActualThing:
public virtual void takeDamage(float damageAmount){
     //reduce hp
//in Sensor:
public override void takeDamage(float damageAmount){
     //reduce hp
     //reduce sight distance

This is how basic object inheritence patterns can be implemented in Unity.

Systems

The Grid: The first programming challenge was an infinite, non-repeating grid. Based on my previous thinking on large pseudorandom world generation, I worked out a system that places BigTiles (containing a 10×10 grid of normal game Tiles and other content on top of them) from a list of available BigTiles. A given x and y will always generate the same BigTile, allowing the game to dispose of BigTiles it no longer needs, but generate them again if needed. Each game picks a random x and y offset to the starting position–a pair of C# ints of value -2 billion to 2 billion. The x and y offsets are summed and used as another pseudorandom seed to further shuffle things around atop the BigTile. The list of available BigTiles changes based on the distance from the game’s starting point, allowing the game to gate more powerful and challenging content. Specific BigTiles can also be forced to appear, like the starting location.

Lots of stuff, not-quite-randomly generated

Pathfinding: The bots use an A* pathfinding system. Niek worked out the initial code, and I adapted it to allow planning without executing the route (for AI reasoning) and to talk to the existing grid-based systems. This required a deep dive into Amit Patel’s A* Pages, a deep summation of pathfinding systems, which I highly recommend.

AI: The bots not being driven by the player have competing desires, to which they assign weights based on need and availability of a solution. The highest weight wins. They may reevaluate their options several times on the way to their goal, but the current “plan” has a sunk-cost-fallacy bonus attached, to reduce indecisiveness. The AI can query the Pathfinder (a separate script) for the “cost” of a path; it will also store the steps needed for pathfinding, to avoid a processing-heavy path recalculation for the action it ultimately decides to take.

I grew up calling them that; turns out they’re “Touch-Me-Nots”

Mystery Boxes: Remember how every BigTile has its own pseudorandom seed number? MysteryBoxes are a system that uses these to “randomly” shuffle things around on BigTiles when they’re generated–plants, upgrades, whatever you’d like. The plant growing on top of a toppled monument? That’s not scripted. One limitation is that if something on a BigTile gets destroyed, it’ll reappear if the player ever goes far enough away and comes back. A special subclass, the CPUBox, will generate a new CPU (the base of a new bot) if your party is smaller than the allowed size, or a random Upgrade if not. Like other gated content, the maximum number of party members increases with distance from your starting point.

Class flowchart. Always out of date.

Tiny Convoy: Plan

“Tiny Convoy” is as a demo-length convoy simulator–a moving base builder–developed in Unity 2019 by Niek Meffert, Lucas Oliveira and myself. Playable betas for Mac and Windows can be downloaded here. The goal is to survive as a group of tiny robots in a big dangerous world: find power, upgrade your bots, and avoid dangers.

Remember the annoying escort missions in “Warcraft III,” where your soldiers kept hauling off to attack anything that moved, instead of protecting the things they were supposed to? It’s like that. Except your bots can’t attack anything. You are essentially robot herbivores. Looked at another way, “Tiny Convoy” is a herd simulator.

One-pager

As a small team in our first year Masters, the work was loosly siloed. Oliveira took the lead on modelling and texturing, Meffert was point on UI and the first version of the A* pathfinder, and the C# code base is about 90% mine. (I also designed the Touch-Me-Not exploding plant.)

1. Emergent visual storytelling–no dialogue, no scripted beats
2. Keep moving & exploring
3. No weapons
4. Cooperation is essential to survive & thrive

Design Pilars

Our goal was to marry emergence and progression in an interesting and satisfying way. We wanted to build a game in which the player could create in their own mind an emergent storyline around an algorithmically-generated world. If the content one encountered became too boring and undifferentiated, there would be no sense of progression, and little incentive to keep exploring. On the other hand, if too many things were deliberately laid in the player’s path at defined points, it would take on the feeling of an “on rails” story game, to the detriment of replayability.

Concept sketch

Every “Civilization” player remembers when the Mongols took their capital, and they rallied back to win. What happened was only the randomized game content interacting with the rules and the player’s choices, but it took on the quality of a narrative moment because it was different and valued by the player. Likewise, we wished to create moments like “When Little Blue got stepped on,” and  “When we found the field of Touch-Me-Nots on the other side of the desert”—moments created entirely by the game systems interacting with the player’s choices, but which would feel personally meaningful.

The game loop for Upgrade Mode

 To do this, we needed to construct a series of game loops, with systems of player feedback. We would gradually challenge the player by introducing limits on stock (not enough power available) and new sinks (dangers that damage parts). The game engine would introduce new content and changes to the starting conditions in a measured manner as the player moved farther and farther across the game world.