English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Intelligent biological systems are characterized by their embodiment in acomplex environment and the intimate interplay between their nervous systemsand the nonlinear mechanical properties of their bodies. This coordination, inwhich the dynamics of the motor system co-evolved to reduce the computationalburden on the brain, is referred to as ``mechanical intelligence'' or``morphological computation''. In this work, we seek to develop machinelearning analogs of this process, in which we jointly learn the morphology ofcomplex nonlinear elastic solids along with a deep neural network to controlit. By using a specialized differentiable simulator of elastic mechanicscoupled to conventional deep learning architectures -- which we refer to asneuromechanical autoencoders -- we are able to learn to perform morphologicalcomputation via gradient descent. Key to our approach is the use of mechanicalmetamaterials -- cellular solids, in particular -- as the morphologicalsubstrate. Just as deep neural networks provide flexible andmassively-parametric function approximators for perceptual and control tasks,cellular solid metamaterials are promising as a rich and learnable space forapproximating a variety of actuation tasks. In this work we take advantage ofthese complementary computational concepts to co-design materials and neuralnetwork controls to achieve nonintuitive mechanical behavior. We demonstrate insimulation how it is possible to achieve translation, rotation, and shapematching, as well as a ``digital MNIST'' task. We additionally manufacture andevaluate one of the designs to verify its real-world behavior.

Neuromechanical Autoencoders: Learning to Couple Elastic and Neural Network Nonlinearity