Abstract
One key challenge in Evolutionary Robotics (ER) is to evolve morphology and controllers of robots. Most experiments in the field converge rapidly to a single solution for the entire population. Early convergence results in a premature loss of diversity, which creates inconsistent results across multiple runs, sometimes converging to a local optimum. In Nature we can observe the opposite behavior: the more time passes, the more life becomes increasingly diverse. The increasing diversity is correlated to the formation of new species, which is catalyzed by reproductive isolation caused by physical or behavioral separation. Inspired by natural evolution, in this paper we apply artificial speciation based on morphological traits to an ER system. Individuals are forced to crossover only with individuals within the same species and a protection mechanism is applied to newly created species. In our experiments, we demonstrate that this speciation mechanism, inspired by NEAT, can evolve a population rich of many coexisting individuals, differing both in morphology and behavior.
Original language | English |
---|---|
Title of host publication | 2020 IEEE Symposium Series on Computational Intelligence (SSCI) |
Publisher | Institute of Electrical and Electronics Engineers Inc. |
Pages | 2272-2279 |
Number of pages | 8 |
ISBN (Electronic) | 9781728125473 |
DOIs | |
Publication status | Published - 5 Jan 2021 |
Event | 2020 IEEE Symposium Series on Computational Intelligence, SSCI 2020 - Virtual, Canberra, Australia Duration: 1 Dec 2020 → 4 Dec 2020 |
Conference
Conference | 2020 IEEE Symposium Series on Computational Intelligence, SSCI 2020 |
---|---|
Country/Territory | Australia |
City | Virtual, Canberra |
Period | 1/12/20 → 4/12/20 |
Keywords
- evolutionary computing
- evolutionary robotics
- robotics
- species