RL with an evolved reward function.
First described in [[Ackley and Littman, 1991 - Interactions between learning and evolution]].
Used as the basis for [[Ashley 2019 - Learning to Select Mates]]
—
What can be learned by trial and error and learning what actions work in what circumstances?
-
Has there been any consideration of MARL for ERL?
- speciation via niches of the objective function?
- Evolve a population of agents that exist in a world
- Agents have a genome that specifies their objective function
- There has to be room for different strategies
- first, try classic game theory examples, e.g. prisioner’s delimena
- **How to build enviroments that have niches?**
-