Combining Model-Based and Statistics-Based Techniques in Cyber-Physical System Design

Abstract

Research will be performed to investigate novel techniques for combining statistics-based and model-based approaches for cyber-physical system design. The proposed work includes three threads of research: (1) Collaboration schema design, (2) Skill acquisition, (3) Support for cyberphysical system programming and engineering. In thread (1), work will be performed to formalize the concept of collaboration schema that integrates the modeled-based and the statistics-based approach for real-time decision making and control. We shall investigate different ways for agents to collaborate by sharing sensor-acquired and computationally derived state information. The interaction between agent(networks) will be characterized by fixed point computations that have both symbolic and numerical elements. We shall identify appropriate metrics for measuring the performance of collaboration schemas including the response time of an agent (network) which may be defined in terms of the time it takes for a recursive computation to reach a fixed point.While the determination of response time is likely to be computationally intractable in most practical settings, we shall investigate ways to estimate response time of agent (network) analytically if possible and also experimentally with measurement tools. Based on the understanding of agent (network) behavior, we shall also look at synthesis methods that are useful in engineering agent networks to meet given performance requirements. In thread (2), work will be performed to investigate effective ways to enable agent (network) to acquire new skills. We are especially interested inunderstanding whether model-based symbolic methods can be exploited to partition the configuration space of a highdimensional system such as a robot so that statistic-based techniques can be more effectively used to train the robot to acquire complex skills, subject to the symbolic constraints. Conversely, we shall also investigate if statistics-based methods can be used to direct the search in symbolic space for solutions in skill acquisition. An intriguing fundamental issue that we shall investigate is the relative importance of the topology of a neural network versus the weight assignment to the edges of the neural network that makes the network effective in accomplishing the task for which it isbuilt. In the context of model-based versus statisticsbased methods, we can think of the edge weight assignment process in each evolution step as a statistics-based computation whereas the addition/removal of a node or edge from the network topology as a symbolic or model-based computation. In this context, we shall try to better understand whenneuro-evolutionary techniques such as NEAT will work and when they will not. In turn, the insight so obtained will be used to guide how best to combine statistics based and model-based methods in search problems for specific application domains. In thread (3), work will be performed to integrate our research results into experimental application platforms such as an autonomous model car or a manufacturing robot for the purpose of system design, implementation and testing. We shall leverage existing open-source software such as ROS (Robot Operating System) to build a suite of tools to enable fast prototyping of our research ideas into applications.

Document Details

Document Type
DoD Grant Award
Publication Date
Feb 03, 2017
Source ID
N000141712216

Entities

People

  • Aloysius Mok

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Texas at Austin

Tags

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms
  • Autonomy
  • Autonomy - Autonomous System Control
  • Cyber
  • Space