Genetic representations in the optimization of three-dimensional agents















Optimization of 3D agents is a challenging task due to a number of reasons. The sources of difficulty include infinite, discrete-continuous search space, solutions containing variable amounts of information, strong dependencies between parts of a solution, multiple local optima, numerous constraints, and the coevolution of bodies (structures, designs) and their controllers (e.g., neural networks). This talk focuses on the key aspect that influences the efficiency of optimization, namely the genetic representation and its accompanying reconfiguration operators. I will demonstrate examples of 3D agents in various real-life applications, describe a common model used to compare various genetic encodings, and show sample results of evolutionary experiments. Finally, I will show a high-level interface to the Framsticks environment that facilitates experimentation, testing, and comparisons of different optimization algorithms in programming languages like Python.


Time&Place  2022.12.01  16: 05,  Zoom






Tutorial length

1 hour







Tutorial level