Citizens

In Banished, the player never controls a citizen directly. Each citizen decides what to do next: eat, get warm, fetch firewood for the home, go to work, visit the doctor. This page explains how that decision is made, and how it becomes walking and working. The code is in crates/game/src/behaviour/ and crates/game/src/people/, and path finding in banished-path.

Three components

A citizen is an entity made from the template Template/Citizen.rsc. Three of its components work together:

Component	What it does
CitizenComponent	The needs (hunger, cold, health, happiness), the age and the household
CitizenAIComponent	Chooses the next activity
CharacterComponent	Runs a stack of actions: walks, turns, plays animations, carries things

Choosing an activity

When a citizen has nothing to do, or when hunger or cold becomes urgent, the AI builds a list of candidate activities, each with a priority. For example:

Activity	When it is a candidate
Eat	Hungry; urgent when very hungry
Get warm	Cold; urgent when very cold
Emergency work	A fire nearby, for example
Fetch clothing	The citizen has no coat
Fetch a tool	An adult without a tool
Stock the home with food or firewood	The home is short of it
Visit the doctor, fetch herbs	Sick, or in poor health
Go to school	A child, if there is a school
Work	An adult with a job
Drop off what it carries	Carrying goods for a long time
Wander	Unhappy for a long time, or nothing else to do

The list is shuffled with the game's random numbers, then sorted by priority, highest first. Activities with the same priority keep their shuffled order. The AI then tries each activity in turn, until one of them gives the citizen something to do. Each activity has a cooldown, so a citizen does not try the same failed thing again at once.

New citizens have no tool and no coat. So the first choices of an adult are: fetch a tool, fetch a coat, work, and wander.

The shuffle uses the shared game random numbers, and the sort is the C runtime's unstable qsort. Both are reproduced exactly, because the order decides which citizen does what (see Exact maths and random numbers).

Actions

An activity becomes a stack of actions for the CharacterComponent. The actions are small steps, for example:

MoveTo, PlayAnimation, PerformWork, PickupRawMaterial, PickupStoredRawMaterial, DropOffRawMaterial, ConsumeRawMaterial, CreateRawMaterial, CompleteTask, CompleteRoad, HuntAnimal, RemoveNaturalResource, ShowTool.

"Fetch firewood for the home" becomes: walk to a storage that has firewood, pick it up, walk home, drop it off. ShowTool puts the right tool in the citizen's hand: an axe, a pick, a scythe, a hoe, a hammer, a fishing rod, a saw, a bucket, a barrow, or the goods it carries.

Walking

MoveTo asks the pathfinder for a path of cells. The pathfinder is an A* search over the map cells, with a cost for each type of cell. For citizens, the costs are 3 to 16: ground costs 4, and some of the cell types that buildings and roads make cost 3. A diagonal step costs 1.4 times a straight one. Water and high ground block walking. Before it searches, the game checks that both ends are in the same connected region, so it does not search for a path that cannot exist.

The citizen then follows the path. Each iteration, it turns towards the next point, moves by the root motion of its walk animation, and is placed on the ground height.

To choose the nearest storage, job or resource, the game does not run a full search for every candidate. It uses a coarser sector graph: the map is cut into 8 × 8-cell sectors, each sector's walkable areas become nodes, and a quick search over the nodes gives a "path rank" distance. These distances are cached for 120 seconds of game time.

Lives

Citizens are born, grow up, go to school, work, form households, have children, grow old and die. They need food of several kinds, firewood and clothing, and their health and happiness depend on them. These rules are in crates/game/src/people/; see CITIZEN-LIFE.md and CITIZEN-NEEDS-AND-HOUSEHOLDS.md.

How it was proved

The region labels, the A* search and the candidate sort were run in the emulator with many random inputs and compared with the port: 20 random maps for the regions, 600 searches for A*, and 400 candidate lists for the sort. All results are identical. The tests are regions_emu, astar_emu and candidate_sort in crates/game/tests/.

Read more

• CITIZEN-AI.md: the decision, the action stack and walking, in detail.
• JOBS.md: construction, work and professions.
• COMMUTE.md: how the town moves workers closer to their work.
• IMMIGRATION.md: new citizens and nomads.