Skip to content

Instantly share code, notes, and snippets.

Created July 14, 2019 08:50
Show Gist options
  • Save SoptikHa2/6ddeee23a37933135cde4df00ee6d831 to your computer and use it in GitHub Desktop.
Save SoptikHa2/6ddeee23a37933135cde4df00ee6d831 to your computer and use it in GitHub Desktop.
Code wars - behaviour tree
// Petr Šťastný <>
// 2019-07-14
// trait = interface
pub trait CodingBlock {
fn execute(&self) -> Option<&Action>;
// Define enums. Each enum variant can
// have another type inside it.
#[derive(Debug)] // Allow debug printing this enum
#[derive(PartialEq)] // Allow comparing this enum
pub enum Action {
pub enum Direction {
pub enum RobotSelector {
pub enum RobotRelationship {
// Define query enums
pub enum Query {
LowerThan(i32, Property),
HigherThan(i32, Property),
pub enum Property {
// Define Action coding block
// This coding block returns an action
// to be done.
pub struct ActionCodingBlock<'a> {
action: &'a Action
impl CodingBlock for ActionCodingBlock<'_> {
fn execute(&self) -> Option<&Action> {
impl ActionCodingBlock<'_> {
fn new(action: &Action) -> ActionCodingBlock {
ActionCodingBlock {
// Define query coding block.
pub struct QueryCodingBlock<'a> {
query: &'a Query,
child: Option<&'a CodingBlock>, // QueryCodingBlock has one child, that gets executed if query is true
impl CodingBlock for QueryCodingBlock<'_> {
fn execute(&self) -> Option<&Action> {
if let Some(child) = self.child { // if child exists
// TODO: Test query and MAYBE execute child
return child.execute();
// We can't return any action
return None;
impl<'a> QueryCodingBlock<'a> {
fn new(query: &'a Query) -> QueryCodingBlock<'a> {
QueryCodingBlock {
child: None
fn register_child(&mut self, child: &'a CodingBlock) {
self.child = Some(child)
// Define hub. Hub connects to multiple coding blocks.
// It tries to execute them from left to right,
// and the first coding block to return an action
// is actually executed.
pub struct HubCodingBlock<'a> {
children: Vec<&'a CodingBlock>
impl CodingBlock for HubCodingBlock<'_> {
fn execute(&self) -> Option<&Action> {
for child in &self.children {
if let Some(action) = child.execute() {
return Some(action);
return None;
impl<'a> HubCodingBlock<'a> {
fn new() -> HubCodingBlock<'a> {
HubCodingBlock {
children: Vec::new()
fn add_child(&mut self, child: &'a CodingBlock) {
fn main() {
// Construct the coding blocks
let move_to_closest_enemy = Action::Move(Direction::Towards(RobotSelector::Closest(RobotRelationship::Enemy)));
let shoot_weakest_enemy = Action::Shoot(RobotSelector::LeastHp(RobotRelationship::Enemy));
let if_hp_below_twenty = Query::LowerThan(20, Property::Health(RobotSelector::ThisRobot));
let run_away_from_closest_enemy = Action::Move(Direction::AwayFrom(RobotSelector::Closest(RobotRelationship::Enemy)));
let if_closest_enemy_50m_or_closer = Query::LowerThan(50, Property::Distance(RobotSelector::Closest(RobotRelationship::Enemy)));
// Construct decision tree
let mut root = HubCodingBlock::new();
let mut query_hp_below_twenty = QueryCodingBlock::new(&if_hp_below_twenty);
let mut query_enemy_close_enough = QueryCodingBlock::new(&if_closest_enemy_50m_or_closer);
let action_run_away_from_enemy = ActionCodingBlock::new(&run_away_from_closest_enemy);
let action_shoot_enemy = ActionCodingBlock::new(&shoot_weakest_enemy);
let action_go_towards_enemy = ActionCodingBlock::new(&move_to_closest_enemy);
// Nodes in decision tree will be executed from left to right (just like going thr. binary tree)
// First action that is found is executed. Actions below queries don't get executed if query fails.
// Because of how Rust works with memory, register queries first
// Root node will contain immutable reference to each node. We need guarantee that
// nodes to which Root node has references will never change. So we must give queries reference to children
// before connecting them to root node.
query_hp_below_twenty.register_child(&action_run_away_from_enemy); // What happens when query is successful?
// Now we can start connecting queries (which will never change at this point) to root node
root.add_child(&query_hp_below_twenty); // hp_below_twenty is most important, so it goes first
root.add_child(&query_enemy_close_enough); // if we have enough HP, check if there is an enemy close enough
root.add_child(&action_go_towards_enemy); // If we didn't run away and we didn't shoot, go towards closest enemy.
// Now we can execute the tree. If there was runtime, we could execute the action. Since
// queries are not evaluated and always return true, we should get &run_away_from_closest_enemy,
// since it's action with highest priority.
// Test if this decision tree returns an action (.unwrap will cause the assert to fail if no value is returned)
// and if the action is run_away_from_closest_enemy.
assert_eq!(root.execute().unwrap(), &run_away_from_closest_enemy);
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment