For patients with amyotrophic lateral sclerosis (ALS), disease progression can cause a loss of motor function. As motor function declines, the dexterity needed to control a wheelchair’s joysticks can also be compromised. The objective of this work is to integrate user sensor inputs and wheelchair position measurements to improve the performance of wheelchair guidance, even in the presence of noisy inputs from the user. This work evaluates probabilistic, model-based methods for blending joystick and position inputs along a series of user-created trajectories, similar to those that a wheelchair user may follow in their day-to-day navigational routines. We answer three key questions in order to associate joystick inputs to path-keeping decisions: 1) What is a path? 2) When are paths different? 3) What is the probability of a particular destination along a path? The algorithmic answers to these questions are verified using experimental wheelchair joystick and position measurements. Using this approach, the goal is to safely guide a wheelchair’s trajectory even if the user is providing ambiguous inputs. This process enables better discrimination of user joystick inputs for navigation algorithms, resulting in improved wheelchair guidance, safety, and patient monitoring.