Date of Graduation
Statler College of Engineering and Mineral Resources
Mechanical and Aerospace Engineering
Practical decision makers are inherently limited by computational and memory resources as well as the time available in which to make decisions. To cope with these limitations, humans actively seek methods which limit their resource demands by exploiting structure within the environment and exploiting a coupling between their sensing and actuation to form heuristics for fast decision-making. To date, such behavior has not been replicated in artificial agents. This research explores how heuristics may be incorporated into the decision-making process to quickly make high-quality decisions through the analysis of a prominent case study: the outfielder problem. In the outfielder problem, a fielder is required to intercept balls traveling in ballistic trajectories, while the motion of the fielder is constrained to the ground plane. In order to maximize the probability of interception, the agent must make good, yet timely, decisions. Researchers have put forth several heuristic approaches to describe how a fielder may decide how to run based only on immediately available information under different control paradigms. This research statistically quantifies upper bounds on the expected catch rate of a couple notable approaches, given that interception of the ball is theoretically possible if the fielder ran directly towards the landing spot with maximal effort throughout the entire duration of the ball’s flight.
Additionally, novel modifications are made to a belief-space variant of iterative Linear Quadratic Gaussian (iLQG), which is an online method that may be used to find locally-optimal policies to continuous Partially Observable Markov Decision Processes (POMDPs) in which Bayesian estimation may reasonably be approximated by an Extended Kalman Filter (EKF). Directional derivatives are used to reduce the computation time of certain matrix derivatives with respect to the variance of the belief state from to , where is the dimension of the belief space. However, the improved algorithm still may not be capable of real-time decision-making by the standards of modern-day computing on mobile platforms, especially in systems with long planning horizons and sparse rewards. The belief-space variant of iLQG is applied to the outfielder problem, which may also indicate its applicability to similar target interception problems with input constraints, such as missile defense.
Lassak, Kyle Gabriel, "Fast Decision-making under Time and Resource Constraints" (2020). Graduate Theses, Dissertations, and Problem Reports. 7521.