Research
Learning to move via Artificial Intelligence
There is a growing effort to understand decentralized control mechanisms, particularly in application to autonomous robotic systems with distributed sensing and actuation. Sea stars, being equipped with hundreds of tube feet, are a model system for this problem. Inspired by our experimental observations, we develop a mathematical model that captures salient features of the sea star biomechanics and nervous systems. We then use Reinforcement Learning (RL) algorithms to train optimal decentralized locomotion controllers and the most effective mechano-sensory cues for locomotion control.Â
In another collaborative project with researchers from DeepMind, we have used RL algorithms to train a 3-link fish model to swim in potential flow. You can read the article here.
Hydrodynamic advantages of swimming in groups
I'm also interested in fluid-structure interactions and understanding how flying birds and swimming fish are able to utilize the strong interaction between their bodies and the surrounding fluid to achieve highly-controlled locomotion. I study the emergent dynamics and coordination in large groups of interacting swimmers using a hierarchy of fluid-structure interactions models. Our results corroborate recent experimental findings on pairs of swimmers and underline the role of hydrodynamic interactions in the collective behavior of swimmers.
Drop formation and capillary flows
I'm also interested in using Computational Fluid Dynamics (CFD) to model multiphase flows. Specifically, I have studied drop formations and capillary flows in the context of Volume of Fluid (VOF) method. Some of my simulations can be found on this Github repository.