Echo state property and memory in octopus-inspired soft robotic arm

Abstract

Soft body generates diverse dynamics that can be used as a computational resource based on the framework of physical reservoir computing. In this study, we set up an experimental platform for a soft continuum arm to extract the skeletal dynamics from its arm motion and to systematically investigate the performance of computational capability. Our focus is to clarify how the length of the soft arm and the environmental condition affect overall information processing capability. First, we collected the video data of the soft arm with different lengths in the air, tap- or sea-water conditions and tracked the arm's centerline motion. Next, we assigned a short-term memory task to the soft arm system and examined how the length of the arm affects these task performances. In addition to examining the memory capacity, we examined the local memory capacity along the arm to quantify the heterogeneity of the capacity. Finally, we examined the echo state property of the soft arm in different environmental conditions by quantifying the differences of two arm responses to the common input sequence. We found that the memory capacity was higher in water than air, but no difference was observed in sea- and tap-water conditions. The local memory capacity showed that the capacity of the tip region was especially low in air. In this condition, the responses of the tip region also did not synchronize as response to the common input. These results suggest that there is weak echo state property in the tip region of the soft arm: the states of the tip are not purely the function of the historical input. Our results demonstrate one of the concise examples on how the physical parameter in body and environment affects the information processing capability of a system.