Vision Based Recognition Using Deep Reinforcement Learning for Unmanned Systems

Abstract

Research on unmanned vehicles that can discover an unknown environment itself has attracted theeffort of researchers around the world for over two decades. The goal of such studies is to makeunmanned vehicles/robots to be able to learn from their environment and navigate themselves to the target rather than training on supervised labels provided by human beings.In this research proposal, we will investigate and propose deep reinforcement learning methods fornavigating unmanned vehicles based on visual information such as RGB color image stream anddepth image stream. First, we will study effective methods to learn to memorize visual informationwhich allows the unmanned vehicle to explore and remember visual structures and properties ofenvironment. Secondly, we will conduct research on deep reinforcement learning algorithms torecognize target objects in the environment and navigate vehicle to the location of particular targetobjects. And lastly, we will explore deep reinforcement learning algorithms to detect unidentifiedobjects that are not seen previously in the environment and navigate vehicles to the location of suchobjects. Our approach to the problem of visual structure memorization is to combine the strength ofConvolutional Neural Networks (CNN) in encoding visual pattern with the sequence modelingpower of Long Short Term Memory (LSTM). The combined system would encode the visualinformation stream generated by the vehicle s movement into effective feature vectors which will bethe input to the movement policy component of the vehicle’s brain. In addition to visual streamencoding, the movement policy component also relies on unified target representation in knowntarget and unknown target situations. This allows the vehicle to navigate through the environment tofind targets and detect unknown objects along its path.

Document Details

Document Type
DoD Grant Award
Publication Date
Apr 09, 2018
Source ID
FA23861714053

Entities

People

  • Le Thanh Ha

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Engineering and Technology

Tags

Fields of Study

  • Computer science

Readers

  • Aerial Unmanned Vehicle Swarm Micro Periodontal Dentistry.
  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Neural Network Machine Learning.

Technology Areas

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