Overview

Project: Mothership Mayhem

Estimated Development Time: 140hrs

Lines Of Code: 3,100 (Including comments and spacing)

Description: The AI Director is in charge of keeping track of the current status of the room(zone). Depending on this status, the Director will change individual enemy's tactics and objectives. The end result is to hopefully keep the gameplay within a single room varied and balanced.

Objectives:

1. Track the different objectives within a given room.

2. Handle the spawning of enemy units.

3. Handle the death and items dropped by enemy units.

4. Distribute objectives to enemy units.

5. Change enemy tactics depending on events and current circumstances.

6. Manage enemy movement (steering behaviors)

Thought Process and Choices

Why I chose to use an AI Director:

Deciding to use an AI Director in Mothership Mayhem wasn't a hard choice. Most of the player's time is spent fighting a wave based system in these randomized room, so it had a natural fit with the type of gameplay we were looking for: the room vs the player. The room(all the enemies as a whole) needed to adapt to the different events occurring. Yes, I could have had enemies communicate to each what is going on and had them adapt on an individual level, but I had a feeling that could get messy very quickly.

An AI Director also made it easier to do more complicated behaviors such as formations and swarms.

Risks:

1. An AI Director gets very big very fast. With all these different moving parts in a single system, it is bug prone and can sometimes take longer than it should to fix a simple issue.

2. Although adding new tactics and strategies to an AI manager is relatively easy, adding new enemies with a whole new set of rules is fairly time consuming. This means for a short project we are limited to only a few enemies with a lot of different uses.

Results:

As risk number two points out, adding new enemies was fairly inefficient. There had been 2 other enemy types in the design process. We instead integrated their abilities and behaviors into already existing enemies. Outside of that, however, the AI Director went fairly smoothly, being built up slowly through the entire development process.

Design and Development

The easiest way to think of an AI Director is to imagine a RTS(real-time strategy) game. The units doing all the fighting in an RTS are the different robots in Mothership Mayhem. The commander in an rts is the AI Director. The different plays and actions made in a RTS are considered tactics for the AI Director. In the end this creates very simple NPCs that merely know how to move and attack. It is the AI Director who tells these units where to move and what to attack.

For this design to work there are 3 main stages that must be in place: Tracking, Movement Coordination, and Tactic Handling. Each of these stages can have many smaller substages within them, all of which are fairly large topics. As you can see at the top, the AI Director has more then 3,000 lines of code. So, for the sake of brevity I am only going through these stages for two systems: herd management and collision avoidance.

Herd Management

A herd is a group of npcs, typically Light Bots, that are trailing behind the player. This was an area of feedback we received early in development from testers and our producers. Basically it made the gameplay into a long back-peddle, and as long the player kept moving they were rarely caught off guard. Herd management spots these trailing groups and performs a few different tactics in order to change up the gameplay.

Tracking:

When dealing with herds, the first thing is to discover whether or not there is even a herd currently active. We do this by taking the average location of all the enemies in a given room. These enemies must be actively pursuing the player in order to be apart of the calculations.

Next, we have to remove the npcs that are too far away from the average position. After removing all the outer npcs, we recalculate the average position.

We repeat this process, making the circle smaller and smaller until we have an accurate description for the herd. Several other checks occur during this process, such as minimal enemy count and distance from the player. But overall, we just want to see where the biggest grouping of enemies currently is.

Once we discovered if there is a herd and which enemies are apart of it, we store this data into a list of known herds. This list is dealt with then latter on in the AI Director.

Movement Coordination:

Having data on a herds center position and all the members apart of it, the AI Director takes a few early counter measures without having to perform any major tactics. One of these is simply sending some of the bots on the outer edge on a looping path. Enemies towards the center of herd are given directions to rise up and dart down in an attempt to get behind the player. This typically leaves 30-40% of the original herd to pursue the player.

In this first video, all herd management systems are turned off. After a few seconds the player eventually has a large numbers of enemies trailing behind him. Although, god mode is turned on for the purposes of testing, you can see that the player simply has to backpedal to avoid most hits.

In this second video, basic movement coordination is enabled. The dart tactic is also being utilized when a herd is being dispersed.

Tactic Handling:

The last aspect managing herds is to change up the gameplay in a large way temporarily. We do this by using tactics. In the video below, the 'Swarm' tactic is being used. Basically, all of the light bots currently in the room start to zip around a single location, like a swarm of flies over some trash. At the end of the swarm, the light bots become still then dart down at the player all at once.

Collision Avoidance

Tracking:

This detection is something that happens last second, a knee jerk reaction to a possible entity-to-entiy collision(the environment is not considered in this check). I do it this way due to the large number of enemies in small areas and the robotic appearance of the enemies themselves. Mothership Mayhem is chaotic by design.

Similar to spotting herds, the AI Director checks enemies that are within each others personal space(an area extending out from an enemy that is roughly 5 times the radius of their collision box). Next, the heading of both enemies are checked. If both enemies are heading towards each other then we simply use the first enemy for the reaction. Otherwise we use the enemy whose heading has a potential collision.

Movement Coordination:

Once a possible collision has been detected, the AI Director figures out which way the npc needs to turn in order to avoid it. The director always bases it decision on how much effort would be required. Meaning if the npc is already turned slightly to the right of the npc ahead of it, then it will continue to turn right. In the off chance it is directly in the center it turns right by default. Then the director simply tells the npc that it needs to turn in a certain direction. If there is still a potential collision a few moments later the process is repeated.

In this first video, all collision avoidance is disabled. The enemies are attempting to surround the player but become stuck in each others' paths.

Collision avoidance is enabled once more in this second video. As you can see, no measures are taken until right before the collision is about to happen. This does mean enemies often rube shoulders, but they eventually make it to their destination.

Postmortem

Final Product:

The AI Director became a massive beast(code-wise) in the end, but it did its job. Gameplay was adjusted based on what was happening in the room. Adding new tactics and behaviors were easy to implement on already existing enemy types. However, adding in new enemy types was a nightmare due to the fact that I would have to go through and add specific conditions along with other checks for each new enemy.

What I Would Have Done Differently:

I have many ideas on how I would have structured it if I had a chance to redo the director, but mostly it comes down to building the system out of more separate modules. A module for all movement. A module for enemy spawning and cleanup. This would help breakup the congestion the AI Director suffered from.