GPU-Computer Cluster System to Enable Deep-Learning Research for Robotic Perception and Motion Planning Algorithms

Abstract

Learning from data is one of the key ways to increase the capability of the next generation ofautonomous robots. While the amount of datasets available to us is large (> 50TB), processingand utilizing this data with commonly available computing resources is not possible. Therefore,we request DURIP funds to purchase a GPU-computer cluster system to enable large-scale deeplearningresearch for robotic perception and motion planning algorithms. The proposed systemconsists of a head node with 88TB of storage, and 17 compute nodes with 4 GPUs and 2 CPUs each.The proposed cluster enables us to push the state of the art in perception and motion planning ofseveral existing ONR and other government sponsored projects. Examples of research areas thatwill be enhanced by the cluster include semantic classification, high-performance motion planning,large scale feature mapping, and image processing algorithms. The proposed system is a catalystthat will amplify our capability to process large datasets and perform large experiments leading tonew advances in autonomy research.

Document Details

Document Type
DoD Grant Award
Publication Date
Sep 30, 2016
Source ID
N000141612903

Entities

People

  • Scott Sanders

Organizations

  • Massachusetts Institute of Technology
  • Office of Naval Research
  • United States Navy

Tags

Fields of Study

  • Computer science

Readers

  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.
  • Research Science/Academic Research

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - Neural Networks
  • Autonomy