Mixed Precision Deep Reinforcement Learning for Continuous Control of Complex Dynamic Systems

Abstract

Deep reinforcement learning (RL) shows promising results for control problems in continuous action spaces. Deep RL is disadvantaged in that it can be very computationally intensive, which is of particular concern when considering fielding deep RL applications on autonomous vehicle platforms with computational and power constraints. Mixed numerical precision methods are an area of active research where progress is being made on improving the computational efficiency of deep learning methods. While mixed-precision approaches are well understood for supervised learning tasks, this area is relatively unexplored for deep RL. The aim of this project is to fill this gap in the research by presenting a method to improve the computational efficiency of the Deep Deterministic Policy Gradient (DDPG) algorithm by using mixed numerical precision and loss scaling. Numerical cases that require continuous control of a complex dynamic system model are presented that quantify the performance and computational improvements of DDPG agents trained with mixed precision as compared with single precision.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 08, 2023
Accession Number
AD1223879

Entities

People

  • Alfa R. Heryudono
  • Christopher J. Hixenbaugh
  • Eugene J. Chabot

Organizations

  • Naval Undersea Warfare Center
  • University of Massachusetts Dartmouth

Tags

Readers

  • Brain and Cognitive Science; Experimental Psychology; Cognitive Neuroscience
  • Robotics and Automation.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks
  • Autonomy
  • Space
  • Space - Spacecraft Maneuvers