@nleadholm explains action policies in Monty, covering model-based learning systems, the interaction between cortical and subcortical structures, the role of abstract spaces in decision-making, and hierarchical goal-directed behavior, with a focus on practical applications and research directions.
Very interesting video about the policies.
I thought about how I observe objects, noises, things outside my norms,… in the world. And most of the time I have the feeling my attention is driven to the point where the maximum in sensational difference is happening between two or more points in time.
Do you guys have any idea that such a action policy like this could be usefull?
Like → Move the sensor towards or around the location where the biggest difference in sensed features was recognized or expected.
Best regardes
Thanks for your question, yes we definitely think something like this will be important! We have some early thoughts written up on this, they will appear in our Readme page on “Top-Down Exploration Policy” soon, but in the meantime you can read about it at this open PR link.
Thank you Niels, I will check it out:)