Summary
Particle swarm optimisation (PSO) is a population-based optimisation algorithm inspired by social search. Each particle remembers its own best position (pbest) and is also pulled toward the best position discovered by the swarm (gbest). Unlike a genetic algorithm, PSO does not rely on crossover and mutation; it updates positions through velocity and memory. (Brabazon, O’Neill and McGarraghy, Natural Computing Algorithms, see source-natural-computing-algorithms)
Key ideas
- Each particle has a position, velocity, personal best, and access to global best.
- Search quality comes from mixing inertia, self-correction, and social influence.
- PSO is often easier to apply to continuous parameter tuning than GA.
In practice
Game uses:
- tuning AI weights
- balancing numeric parameters
- searching level-generator parameters
- evolving controller settings without hand-tuning every variable
Evidence
- The source frames PSO as superior to blind random search because it preserves both memory of past success and social communication across the swarm.
- The particle’s next move depends on previous velocity plus attraction toward
pbestandgbest, which gives the method both momentum and convergence pressure.
Implications
- PSO is often a good fit when the search space is continuous and you are optimising parameters rather than symbolic plans.
- It pairs well with simulation-based evaluation in games, where a candidate can be run and scored repeatedly.
Open questions
- Which game-design tuning problems in this vault would make the clearest PSO teaching example?
- Should PSO eventually get a Unity visual demo similar to the Nature of Code examples?
Related
genetic-algorithms · neuroevolution · machine-learning-games · source-natural-computing-algorithms