Agent-Environment Interaction in Visual Homing

This study illustrates how obstacle avoidance can emerge from a visual homing strategy, caused by the intrinsic geometric structure of the environment. An example is shown where an agent performs visual homing in a virtual environment with several obstacles which also serve as visual landmarks. The agent has omnidirectional vision similar to many prey animals. The applied visual homing strategy is the Average Landmark Vector (ALV) model by Lambrinos et al.[1]. When observing the homing trajectories of the agent, it can be seen that it performs obstacle avoidance without having this behaviour explicitly encoded. It will be shown that the dynamic feedback the agent gets from its environment is crucial for this kind of behaviour.