Mechanics-Based Material Computing using Physical ReLU Spring Networks

Dynamic movement and shape change are key enablers for living systems to sense, assess and respond to environmental stimuli. For example, actuating mechanisms are utilized in natural systems for diverse operations, such as arms and legs for locomotion, skin wrinkling for camouflage, or multistable snapping to catch prey. Nonlinear mechanical dynamics can also serve as a computing resource to directly process the environmental cues into complex mappings between the input and output dynamics of a mechanical system. To explore this concept, we numerically investigated the computing capacity of 2D nonlinear spring networks using a reservoir computing approach. Reservoir computing is a class of recurrent neural networks that trains only a readout layer of the network dynamics in contrast to tuning all the internal parameters of the network. We introduce a mechanical analog for the rectified linear unit (ReLU) activation function from neural network community and benchmark the memory capacity, nonlinearity and output tasks of mechanical ReLU networks sampled from a distribution of spring properties. Preliminary results indicate that the stiffness ratio of the ReLU spring (ie. ratio of the bilinear slopes) is a key driver of the nonlinearity score of the network, even more so than the incidence of activating the spring nonlinearity. In addition, spring networks with a mixture of linear and ReLU springs exhibit a marked loss in memory capacity even at low ReLU springs fractions. Collectively, the results highlight the potential to harness computing capacity from the intrinsic dynamics of a material or mechanical system and motivates further study on how to leverage unconventional materials and physics for information processing.