resetting behavior

Retrieving food from the environment and returning it home is a well known behavior. We are interested how organisms navigate in order to do so, and which evolutionary constrains and factors shape the behavior. During one of our experiments to evolve path integration we observed reseting behavior. Imagine you drive to a new location and suddenly get lost, what do you do? Easy, you return to a known place and try again. This is exactly what bees do, and what our agents also evolved to do. In this example you see a bee (very small) coming from the right and flying to a place to the left outside of the camera range. The bee is trained to know that there is a food source to the left, but in this experiment the food source is hidden from the bee. In the middle is a yellow marker that can only be seen from the right. When you observe is the bee first flying to the left, but once it can’t find the food, it returns to the yellow mark in order to reset and start again.

In the following video you see an virtual agent in black. It starts on his home (green) and needs to go to the yellow area where it finds food. We evolved this agent’s Markov Brain by rewarding agents that retrieve as much food and return home as possible. However, the home field is not constantly visible, but has a chance of showing dependent on the distance of the agent. The further the agent is away the lower the probability to see. The agent has a retina that allows it to see these areas from far away. What you see is that this agent as well as the bee frequently returns to his food place in order to reset.

Our next research goal is to understand under which conditions this behavior evolves, and how the agent performs it.


You may wander what motivates this behavior? If you or the bee can’t find what you are looking for, your expectation gets disappointed, and the emotion you are feeling is frustration. If you think the bee is frustrated and feels this as an emotion, do you think the virtual agent is frustrated too? I would like to argue that we evolved an emotion here, namely frustration. While the intensity is not comparable to what we experience, I still think that the virtual agent is as frustrated as the bee.

Comments are closed.