Strategies for Two Alternative Forced Choice Navigation Tasks

Abstract

In the engineering community, there has been a growing interest in emulating robust continuous decision mechanisms, often found implicitly in biological systems, in engineered autonomous systems. The Two Alternative Forced Choice navigation problem, where an agent is required to decide between two possible navigation tasks based on a noisy signal, is a natural model problem for studying such mechanisms. In this thesis, we aim to generate and systematically study a variety of decision-making strategies in terms of the expected time, distance and error rates. We look at five in particular; the first four are preliminary, based off of various heuristics, while the fifth follows a model predictive control framework with a novel adaptive cost function that penalizes the control magnitude and deviation from a dynamic “artificial” goal. The strategies are studied using a variety of computational and analytic methods; for the model predictive control strategy in particular, closed form results for the expected trajectory in both deterministic and stochastic environments are presented.