An Adaptive Markov Process for Deceptive Robotics

Restricted (Penn State Only)
Author:
Ayub, Ali
Graduate Program:
Electrical Engineering
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
July 31, 2017
Committee Members:
  • Aldo Morales, Thesis Advisor
  • Seth Wolpert, Thesis Advisor
  • Amit Banerjee, Committee Member
Keywords:
  • Deception
  • Robotics
  • Probability Theory
  • HRI
  • Markov Process
  • Entropy
Abstract:
Deception has a long history regarding the study of intelligent systems and it is a common behavior in all of the intelligent beings ranging from insects to humans. Animals and humans both gain several advantages from their deceptive abilities, hence researchers have started to develop different ways to introduce deception in robots. Although deception is really important, not a lot of work has been done on its applications in robotics. In this paper, previous work on deceptive robot motion in a binary target system for a manipulator is analyzed and applied on a mobile robot simulator with some further improvements. Moreover, a new adaptive algorithm is proposed to randomly choose deceptive trajectories, based on the previous choice of them, to deceive humans in the long run. A mathematical formulation of the deceptive algorithm using Markov decision processes (MDP) is also presented and an updated Markov process (called Adaptive Markov Process) is developed. The mathematical model presented is an extension to a simple higher order Markov process using a number of new factors and variables, in particular the inclusion of reset and distribution of probabilities in a more intelligent manner. Different studies were performed with human participants to test deceptive strategies’ performance on a mobile robot simulator. Furthermore, studies also showed that the adaptive deceptive algorithm is more effective in deceiving humans in the long run than just a normal random algorithm. A different set of user studies were performed with human participants to test our adaptive deceptive algorithm and analyze the effectiveness of each deceptive strategy in the long run on a mobile robot. Our studies showed that the adaptive algorithm did deceive humans in the long run and each strategy was either more or equally deceptive when used in the adaptive algorithm than when used independently. Future possibilities in deceptive robot motion are discussed at the end that could lead to the development of an algorithm for deceptive robot motion planning in a multi-target environment.