Syllabus
- Behavior-based robotics, robot learning, artificial life. Engineering perspective, biological perspective.
- Genetic algorithms, artificial neural networks, neural control of a robot, evolution of neural networks, genetic programming. Robot evolution - simulated and physical.
- Evolution of simple navigation - straight motion with obstacle avoidance.
- Reactive intelligence, sensory-motor coordination.
- Modular control architecture, evolution of a modular architecture.
- Learning and evolution - two forms of (biological) adaptation.
- Competitive co-evolution, a predator-prey model.
- Genotype, phenotype, mapping of genotype into phenotype.
- Evolution of complex walking robots.
- Evolutionary learning of neural networks - algorithms SANE, ESP and NEAT.
- Robotic swarms - examples of usage, coordinated exploration, transportation and clustering, reconfigurable robots.
- From simulation to reality - construction of physical robots based on results from a simulation.
Annotation
Evolutionary robotics is a technique of automatic programming of autonomous robots. The lecture shows how robot can be learned to solve tasks instead of their direct programming.
Algorithms simulating natural evolution (mainly genetic algorithms with neural networks) enable the robots to evolve their abilities in interaction with their environment. In the accompanying seminary, the students will work with robot simulators and robotic kits.