Steering behaviors
A set of simple steering forces — seek, flee, arrive, wander, avoid obstacles — combined into one movement vector for an autonomous character.
Craig Reynolds formalised the idea in his GDC paper Steering Behaviors For Autonomous Characters (1999). Each behavior returns a small force: seek pulls toward a target, flee away from a threat, arrive eases to a stop, wander gives random roaming and obstacle avoidance steers clear. The forces are weighted, summed and clamped to the agent's maximum acceleration.
Steering sits beneath a higher decision layer — a behavior tree or state machine picks what to do, the steering behaviors handle how the character gets there. They build directly on Reynolds's own boids model and underpin crowds, enemy AI and traffic simulations, often paired with a navmesh for the large-scale route.