Bevy Steering

This crate aims to provide steering behaviors for autonomous agents in a Bevy game. It is not a complete navigation solution, but fits into a hierarchy of systems that contribute to agent motion.

Status of this crate

This crate is usable but I can't guarantee it's good. I hope to improve it over time, especially documentation and polish.

When to use this crate

Use this crate for:

• Moving an agent.
• Making agents follow things.
• Keeping agents from running into things.
• Making groups of agents move naturally (boids).

This crate cannot provide:

• Navigation and pathfinding.
• Planning complex routes.
• Making decisions based on surroundings.

The best way to make use of this crate is to combine it with other crates and systems to solve larger "planning" problems, and then feed the output of those systems as inputs into a Bevy Steering system. Look at examples/maze.rs to see how to incorporate a pathfinding with PathFollowing and Avoid behaviors.

Movement

Bevy Steering provides two modes of moving:

• OmniDirectional (default). This ignores the direction of an entity and applies linear force to move the agent in any direction. This is ideal for really simple agents that don't have a front or back, or you wish to turn the agent manually.
• Directional. This assumes the entity can only move along the forward/backward vector and aggressively damps velocity in the lateral directions. The agent will turn to face the destination. This is ideal for character models, simple vehicles, etc.

Implementation

The behaviors are implemented using the "Context Steering" approach. Agents move along the most interesting heading, while avoiding dangerous directions. Force is applied to maintain an agents max_velocity.

Behaviors planned / implemented

• Alignment
• Approach
• Avoid (obstacles)
• Cohesion
• Evasion
• Flee
• Path following
• Pursuit
• Seek
• Separation
• Wander

Most of the behaviors are described in this paper: https://www.red3d.com/cwr/papers/1999/gdc99steer.pdf.

However, not all behaviors are implemented as described. It is not a priority to implement all behaviors in this paper.

LLM (Large Language Model) disclaimer

This crate was made with the assistance of LLMs and coding agents. They made it possible to produce this crate by cutting the amount of time needed to implement most of the behaviors. That does not mean this crate is "vibe-coded". Human effort went into making a clean interface. Code polish and tests will continue to be human-driven.

You are free to submit PRs with the use of LLM tools. Please disclose when LLM tools are used in your PR.