Evolving Virtual Creatures

How to read this

Category

Method: evolutionary co-design of creature morphology and control

Contributions

• A genetic-algorithm system that co-evolves 3D creature morphology and neural control together
• Fitness-driven discovery of locomotion behaviors (swimming, walking, light-following) in simulated physics, with no hand-authored bodies or controllers

Context

An early landmark in learned, physics-based procedural animation, joining artificial-life and evolutionary-computation ideas to character generation rather than building on a specific prior graphics paper.

Correctness

Behaviors are demonstrated qualitatively inside a simulated physical world, so plausibility is judged by the simulator and the chosen fitness functions rather than against real organisms; reproducibility depends heavily on those simulation and fitness choices.

Clarity

Conceptually accessible and famous for its visuals; a first pass conveys the idea, a second pass pays off for the genotype encoding and the evolution loop.

How to read it

First pass for the co-evolution concept and the genotype-to-phenotype graph idea; do a second pass only if you care about how morphology and the neural controller are jointly encoded and mutated.