Controlling Soft Robots with Morphological Communication and SNNs

Soft robots are a kind of robot composed of flexible materials, as opposed to "typical" robots made of rigid metal or plastic. The soft robots in use at Union College are composed of hollow silicon cubes that expand and contract in response to hydraulic pressure. The problem with these designs is that they are notoriously difficult to control and develop gaits for by hand. Recently, researchers have explored control methods that embrace nonlinear dynamic coupling instead of suppressing it. Such designs leverage dynamical coupling for communication between different parts of the robot. Morphological communication refers to when those dynamics can be used as an emergent "data bus" to facilitate coordination among independent controller modules within the same robot. Previous research with tensegrity-based robot designs has shown that evolutionary learning models that evolve spiking neural networks (SNN) as robot control mechanisms are effective for controlling non-rigid robots. Our own research explores the emergence of morphological communication in an SNN-based simulated soft robot in the EvoGym environment.