# Steering behaviours

Agents that seek, flee, wander, arrive and flock

```{raw} html
<a class="example-run-button" href="../demos/features_ai_steering.html">▶ Run in browser</a>


```

**Tags:** `ai` `steering` `boids` `numpy`

Forty autonomous agents steer with the classic Reynolds behaviours, all
vectorised in numpy. Number keys 1-4 switch every agent between seek, flee,
wander and arrive; B switches to a boids flock built from separation,
alignment and cohesion. The mouse is the target, and the highlighted agent
draws its per-behaviour steering forces as coloured debug lines.


## What it demonstrates
- Steering as "desired velocity minus current velocity", clamped to a force
  budget, integrated per frame in on_update.
- Whole-population numpy updates: one (N, 2) array each for position and
  velocity, no per-agent Python loop.
- Boids from three pairwise components (separation / alignment / cohesion)
  computed with a single (N, N) distance matrix.
- Immediate-mode debug drawing: force vectors on one highlighted agent, with
  a legend naming each component.

Controls:
  Mouse   - Move the target
  1/2/3/4 - Seek / Flee / Wander / Arrive
  B       - Boids flock
  ESC     - Quit

Run: uv run python examples/features/ai/steering.py
Headless self-check: uv run python examples/features/ai/steering.py --test

## Source

```{literalinclude} ../../examples/features/ai/steering.py
:language: python
:linenos:
```