The earlier topics in this course treat evolutionary computation as optimization: find the best solution to a fixed problem. Much recent work has moved in other directions. Quality-diversity algorithms keep a large archive of solutions that are each the best of their kind, which is useful when the goal is a repertoire of options rather than one answer, and which can avoid traps that defeat a single-minded search. Large language models have become variation operators. In FunSearch and AlphaEvolve, a model proposes edits to programs, and an evolutionary loop with an automated evaluator decides which edits survive. That combination has produced new results in mathematics and computing.
The field's long-standing applications continue alongside these. Evolutionary robotics designs bodies and controllers together, in simulation and occasionally in hardware. Engineering design has a record of evolved artifacts that were built and used, the best known of which is an antenna flown on a NASA spacecraft. Digital evolution has a long record of surprising its designers by finding solutions they did not anticipate, many of which exploit flaws in the simulation or the fitness function. And artificial life studies evolution itself in software, asking what it would take for an evolving system to keep producing novelty without end. This topic surveys each of these areas and closes the course.
Articles
New methods
- Quality diversity: MAP-Elites and novelty search: searching for novelty instead of fitness, archives of diverse high performers, and the robot that adapts to damage, measured on a robotic arm benchmark.
- AlphaEvolve and LLM-guided evolution: FunSearch, AlphaEvolve, and evolution through large models, what each contributed, and what remains evolutionary about them.
Applications
- Evolutionary robotics: evolved virtual creatures, the reality gap, voxel-based soft robots, and the design of bodies and brains together.
- Evolved designs in engineering: the ST5 antenna, evolved hardware, scheduling, and where evolutionary methods are and are not the tool of choice in engineering practice.
Evolution as a subject
- The surprising creativity of digital evolution: solutions that exploit flaws in simulations and fitness functions, their relation to specification gaming in reinforcement learning, and lessons for designing objectives.
- Artificial life platforms: Tierra, Avida, and continuous cellular automata, the evolution of complex features in digital organisms, and open-endedness as a research problem.
Related topics
Quality diversity extends the diversity-preserving methods of genetic algorithms and the archives of multi-objective optimization. LLM-guided evolution continues the program evolution of genetic programming. Evolutionary robotics builds on neuroevolution and coevolution, and the course's final project evolved soft robots in the same tradition.
References
- Lehman, J., et al. (2020). The surprising creativity of digital evolution: a collection of anecdotes from the evolutionary computation and artificial life research communities. Artificial Life, 26(2), 274–306.
- Mouret, J.-B., and Clune, J. (2015). Illuminating search spaces by mapping elites. arXiv:1504.04909.
- Novikov, A., et al. (2025). AlphaEvolve: a coding agent for scientific and algorithmic discovery. arXiv:2506.13131.