More Like this

1. Neuromechanical Autoencoders: Learning to Couple Elastic and Neural Network Nonlinearity

   Oktay, Deniz; Mirramezani, Mehran; Medina, Eder; Adams, Ryan P. (April 2023, International Conference on Learning Representations (ICLR))

   Intelligent biological systems are characterized by their embodiment in a complex environment and the intimate interplay between their nervous systems and the nonlinear mechanical properties of their bodies. This coordination, in which the dynamics of the motor system co-evolved to reduce the computational burden on the brain, is referred to as "mechanical intelligence" or "morphological computation". In this work, we seek to develop machine learning analogs of this process, in which we jointly learn the morphology of complex nonlinear elastic solids along with a deep neural network to control it. By using a specialized differentiable simulator of elastic mechanics coupled to conventional deep learning architectures---which we refer to as neuromechanical autoencoders---we are able to learn to perform morphological computation via gradient descent. Key to our approach is the use of mechanical metamaterials---cellular solids, in particular---as the morphological substrate. Just as deep neural networks provide flexible and massively-parametric function approximators for perceptual and control tasks, cellular solid metamaterials are promising as a rich and learnable space for approximating a variety of actuation tasks. In this work we take advantage of these complementary computational concepts to co-design materials and neural network controls to achieve nonintuitive mechanical behavior. We demonstrate in simulation how it is possible to achieve translation, rotation, and shape matching, as well as a "digital MNIST" task. We additionally manufacture and evaluate one of the designs to verify its real-world behavior. 

   more » « less
2. Neuromechanical Autoencoders: Learning to Couple Elastic and Neural Network Nonlinearity

   Oktay, D.; Mirramezani, M.; Medina, E.; Adams, R. P. (January 2023, International Conference on Learning Representations (ICLR 2023))
3. Specularly reflected whistler: A low-latitude channel to couple lightning energy to the magnetosphere

   https://doi.org/10.1126/sciadv.ado2657  

   Sonwalkar, Vikas S; Reddy, Amani (August 2024, Science Advances)

   Lightning-generated whistlers profoundly affect the energetic particle population in Earth’s radiation belts, influencing space weather and endangering astronauts and satellites. We report the discovery of specularly reflected (SR) whistler in which the lightning energy injected into the ionosphere at low latitudes reaches the magnetosphere after undergoing a specular reflection in the conjugate ionosphere, contradicting previous claims that lightning energy injected at low latitudes cannot escape the ionosphere. SR whistlers provide a low-latitude channel to transport lightning energy to the magnetosphere. We calculate the relative contributions of SR, magnetospherically reflected, subprotonospheric, and ducted whistlers to the lightning energy reaching the magnetosphere. When SR whistlers are considered, the global lightning energy contribution to the magnetosphere doubles, implying that the previous estimates of the impact of lightning energy on radiation belts may need substantial revisions. Whistler dispersion and intensity analyses quantitatively confirm our results and suggest new remote-sensing methods of the magnetosphere, ionosphere, Earth-ionosphere waveguide, and lightning flashes. 

   more » « less
4. XFEM to couple nonlocal micromechanics damage with discrete mode I cohesive fracture

   https://doi.org/10.1016/j.cma.2019.112617  

   Jin, Wencheng; Arson, Chloé (December 2019, Computer Methods in Applied Mechanics and Engineering)

   null (Ed.)
5. Fish couple forecasting with feedback control to chase and capture moving prey

   https://doi.org/10.1098/rspb.2024.1463  

   Martin, Benjamin T; Sparks, David; Hein, Andrew M; Liao, James C (September 2024, Proceedings of the Royal Society B: Biological Sciences)

   Predator–prey interactions are fundamental to ecological and evolutionary dynamics. Yet, predicting the outcome of such interactions—whether predators intercept prey or fail to do so—remains a challenge. An emerging hypothesis holds that interception trajectories of diverse predator species can be described by simple feedback control laws that map sensory inputs to motor outputs. This form of feedback control is widely used in engineered systems but suffers from degraded performance in the presence of processing delays such as those found in biological brains. We tested whether delay-uncompensated feedback control could explain predator pursuit manoeuvres using a novel experimental system to present hunting fish with virtual targets that manoeuvred in ways that push the limits of this type of control. We found that predator behaviour cannot be explained by delay-uncompensated feedback control, but is instead consistent with a pursuit algorithm that combines short-term forecasting of self-motion and prey motion with feedback control. This model predicts both predator interception trajectories and whether predators capture or fail to capture prey on a trial-by-trial basis. Our results demonstrate how animals can combine short-term forecasting with feedback control to generate robust flexible behaviours in the face of significant processing delays. 

   more » « less