Abstract:Computation, mechanics and materials merge in biological systems, which can continually self-optimize through internal adaptivity across length scales, from cytoplasm and biofilms to animal herds. Recent interest in such material-based computation uses the principles of energy minimization, inertia and dissipation to solve optimization problems. Although specific computations can be performed using dynamical systems, current implementations of material computation lack the ability to self-learn. In particular, the inverse problem of designing self-learning mechanical systems which can use physical computations to continuously self-optimize remains poorly understood. Here we introduce the concept of self-learning mechanical circuits, capable of taking mechanical inputs from changing environments and constantly updating their internal state in response, thus representing an entirely mechanical information processing unit. Our circuits are composed of a new mechanical construct: an adaptive directed spring (ADS), which changes its stiffness in a directional manner, enabling neural network-like computations. We provide both a theoretical foundation and experimental realization of these elastic learning units and demonstrate their ability to autonomously uncover patterns hidden in environmental inputs. By implementing computations in an embodied physical manner, the system directly interfaces with its environment, thus broadening the scope of its learning behavior. Our results pave the way towards the construction of energy-harvesting, adaptive materials which can autonomously and continuously sense and self-optimize to gain function in different environments.

Experimental Demonstration of Coupled Learning in Elastic Networks

Experimental demonstration of coupled learning in elastic networks

Neuromechanical Autoencoders: Learning to Couple Elastic and Neural Network Nonlinearity

C^2:Co-design of Robots via Concurrent Networks Coupling Online and Offline Reinforcement Learning

The Physical Effects of Learning

Self-learning mechanical circuits

Physical effects of learning

Coupled Nanomechanical Graphene Resonators: A Promising Platform for Scalable NEMS Networks

Intelligent mechanical metamaterials towards learning static and dynamic behaviors

Realization of programmable nanomechanical lattice with both nearest-neighboring and next-nearest-neighboring couplings

Adaptive nodes enrich nonlinear cooperative learning beyond traditional adaptation by links

Synchronization of Complex Interconnected Neural Networks with Adaptive Coupling

Deep Convolutional Neural Networks to Predict Mutual Coupling Effects in Metasurfaces

Machine Learning Without a Processor: Emergent Learning in a Nonlinear Electronic Metamaterial

Collective Responses in Electrical Activities of Neurons under Field Coupling

Learning unidirectional coupling using echo-state network

Mechanical neural networks: Architected materials that learn behaviors

Coupled online learning as a way to tackle instabilities and biases in neural network parameterizations

Microscopic imprints of learned solutions in adaptive resistor networks

Machine learning without a processor: Emergent learning in a nonlinear analog network

Modeling and Control of Intrinsically Elasticity Coupled Soft-Rigid Robots