University of Southern California Dissertations and Theses

Evaluating sensing and control in underwater animal behaviors

(USC Thesis Other)

Evaluating sensing and control in underwater animal behaviors

PDF

Download

Open document

Flip pages

Download a page range

Copy asset link

Request this asset

Abstract (if available)

Abstract Animal behaviors are largely shaped by the physics of the surrounding environments. In particular, marine animals evolved distinctive control strategies to sense and respond to stimuli in fluid environments for essential activities such as underwater navigation, predation, and evasion. Notably, the complexity and variability of fluid environments make it very challenging to engineer feedback control laws that achieve a similar level of efficacy. The organisms need enough sensory information to take action, but too much sensing can be noisy and energy-inefficient. It is also unclear what type of sensory information should be valued in decision-making. Moreover, there are physical constraints on the availability of sensory cues and the maneuverability of the body.
In this dissertation, I will describe how we combined deep reinforcement learning, biomechanics, and statistical methods to address the interdisciplinary research questions on underwater sensing and control.
I will start with two major building blocks of my research projects. chapter 1 provides an overview of model-free reinforcement learning applied to embodied systems. I will show the mathematical foundations behind the reinforcement learning framework and focus on the imple- mentation of the actor-critic method. I will also discuss how the design choices in RL should be made based on the scope and the physical meaning of the learning problem. In chapter 2 intro- duces a three-link fish model. Starting from rigid body motions in inviscid fluid, I will explain in depth the mathematical formulation of the three-link fish and describe a geometric tool for model interpretation and gaits design. In chapter 3, we use deep reinforcement learning (DRL) to find optimal shape changes for the three-link fish model to swim and turn. With the analytic tools from geometric mechanics, we show that the DRL control policy is more versatile and more robust than our manually designed control rules. When the flow environment is perturbed by a drift, we can run the same training to get a DRL control policy that adapts and takes advantage of the drift. In chapter 4, we use DRL in a more challenging navigation problem, where the navigator needs to reach a target in unsteady adversarial flows. We find the velocity gradient an effective sensory cue to arrive at a successful control policy with egocentric sensing. Then we compare the performance of the egocentric control policy to a geocentric control policy in terms of success rate, efficiency, and generalizability. We also explore the potential of training in a reduced-order flow environment to get a working control policy, which cuts the computational cost and improves interpretability. chapter 5 presents another type of sensory feedback problem where the control decision is made only once. Instead of learning to control strategy, we examine an array of theoretical models. We build a statistical framework that can quantitatively evaluate and compare the performance of multiple behavior models based on experimental measurements, using the example of the fish’s evasion strategy in response to the predator. We show the escape direction of larval zebrafish is best explained by an “orthogonal” strategy, and is compromised by the physical constraint in turning. Finally, in chapter 6, we conclude this dissertation with some ending remarks on my PhD work.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Mechanical and flow interactions facilitate cooperative transport and collective locomotion in animal groups

PDF

Underwater navigation strategies and emergent collective behavior in bioinspired swimmers

PDF

Synchronization in carpets of model cilia: numerical simulations and continuum theory

PDF

Physics-based and data-driven models for bio-inspired flow sensing and motion planning

PDF

Flow physics and nutrients transport in microscopic living systems

PDF

Traveling sea stars: hydrodynamic interactions and radially-symmetric motion strategies for biomimetic robot design

PDF

Transitions in the collective behavior of microswimmers confined to microfluidic chips

PDF

Acceleration of deep reinforcement learning: efficient algorithms and hardware mapping

PDF

Exploiting mechanical properties of bipedal robots for proprioception and learning of walking

PDF

A machine learning based approach to work process knowledge capture and transfer.

PDF

Aerodynamics at low Re: separation, reattachment, and control

PDF

Reward shaping and social learning in self- organizing systems through multi-agent reinforcement learning

PDF

Dynamic modeling and simulation of flapping-wing micro air vehicles

PDF

Reinforcement learning in hybrid electric vehicles (HEVs) / electric vehicles (EVs)

PDF

Novel soft and micro transducers for biologically-inspired robots

PDF

Enhancing collaboration on the edge: communication, scheduling and learning

PDF

Flow and thermal transport at porous interfaces

PDF

Model reduction of multi-scale interactions in periodic and turbulent flows

PDF

Controlling polymer film patterning, morphologies, and chemistry using vapor deposition

PDF

Pattern generation in stratified wakes

Asset Metadata

Creator Jiao, Yusheng (author)

Core Title Evaluating sensing and control in underwater animal behaviors

Contributor Electronically uploaded by the author (provenance)

School Andrew and Erna Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Mechanical Engineering

Degree Conferral Date 2023-05

Publication Date 05/04/2023

Defense Date 03/20/2023

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag animal behavior,bio-inspired motion,Control,fluid mechanics,OAI-PMH Harvest,reinforcement learning,sensing

Format theses (aat)

Language English

Advisor Kanso, Eva (committee chair), Luhar, Mitul (committee member), McHenry, Matt (committee member), Nakano, Aiichiro (committee member)

Creator Email jiaoyush@usc.edu,yusheng9559@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113099728

Unique identifier UC113099728

Identifier etd-JiaoYushen-11778.pdf (filename)

Legacy Identifier etd-JiaoYushen-11778

Document Type Dissertation

Format theses (aat)

Rights Jiao, Yusheng

Internet Media Type application/pdf

Type texts

Source 20230505-usctheses-batch-1037 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu