Analysis of Structure in Feed-Forward Spiking Neural Networks and Applications to Neuromorphic Control of High Degree-Of-Freedom Robotic Manipulators

<a href="https://embed.stanford.edu/iframe/?url=https%3A%2F%2Fpurl.stanford.edu%2Fyk149rh6130" class="su-underline">Show Content</a>

Abstract/Contents

Abstract: Computing task jacobians (differential maps from the joint space to the operational space of a robotic manipulator) through the use of spiking neural networks (SNNs) has significant applications to power-efficient robot control. However, the complexity of such differential maps increases substantially with the dimensionality of their inputs, rending the problem of SNN function-fitting nontrivial for high-DOF redundant manipulators. In this thesis, we present a function factorization scheme and neural architecture that overcomes some of the challenges associated with fitting. We perform systematic structural analysis of SNNs under this architecture, and show that such a scheme can accurately decode complex multidimensional functions. In particular, we show that task jacobians can be decoded accurately over the full (discretized) joint space of a redundant manipulator using the factorized neural architecture. Specifically, we were able to achieve an RMS error of less than 0.03 per component over the joint space of a six degree-of-freedom manipulator. This level of decode accuracy represents substantial improvement over prior work in this area. Because the SNN implementations we developed were based on Izhikevich neuron models, the results of this thesis can potentially be applied to controlling robots accurately using low-power neromorphic chips.

Type of resource	text
Date created	[ca. June 2018]

Author	Sriram, Vinay Kotikalapudi
Degree granting institution	Stanford University, Department of Computer Science
Primary advisor	Oussama Khatib

Subject	Neuromorphic Computing
Subject	Spiking Neural Networks
Subject	Robotics
Subject	Task Space Control
Subject	Jacobian Matrix
Subject	Factorizations
Subject	Discrete Fourier Transform
Subject	Stanford Department of Computer Science
Genre	Thesis

Location	https://purl.stanford.edu/yk149rh6130

Use and reproduction: User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor.
License: This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).

Preferred Citation: Sriram. V, "Analysis of Structure in Feed-Forward Spiking Neural Networks and Applications to Neuromorphic Control of High Degree-Of-Freedom Robotic Manipulators", Stanford Digital Repository, 2018. Available at: https://purl.stanford.edu/yk149rh6130

Undergraduate Theses, School of Engineering

Loading usage metrics...