Mechanical problem solving in mice

Abstract

Recent advances in automated tracking tools have sparked a growing interest in studying naturalistic behavior. Yet, traditional decision-making tasks remain the norm for assessing learning behavior in neuroscience. We introduce an alternative sequential decision-making task for studying mouse behavior. It consists of an open-source, 3D-printed "lockbox", a mechanical riddle that requires four different mechanisms to be solved in sequence to obtain a reward. During the task, the mice move around freely, allowing the expression of complex behavioral patterns. We observed that mice willingly engage in the task and learn to solve it in only a few trials. To analyze how the mice solved the task, we recorded their behavior in a multi-camera setup and developed a custom data analysis pipeline to automatically detect the interactions of the mice with the different lockbox mechanisms for a large corpus of video footage (> 300h, 12 mice). The pipeline allows us to further delineate why mouse performance increases over trials. Our analyses suggest that this is not due to an increased interaction time with the task or the acquisition of a smart solution strategy, but primarily due to habituation to the lockbox. Lockboxes may hence be a promising approach to study both abstract sequential decision making and low-level motor learning in a single task that can be rapidly learned by mice.