Modeling human movement with length-normalized action primitives

Human movement appears constructed of primitives that serve as building blocks for complex motion. Two pioneering descriptions of aimed movement, the deterministic iterative-corrections and stochastic optimized-submovement models, each assume that such primitives (or submovements) are discrete and non-overlapping. Although these prominent models successfully explain the speed-accuracy tradeoff familiarly known as Fitts´ law, they say little about the kinematic shape of submovements. This paper introduces a family of length- normalized action primitives (LNAPs) that produce rigid-body trajectories similar to those seen in human motion, and which agree with Fitts´ law. Like the aforementioned models, the LNAP method requires discrete submovements. However, it extends the antecedent models by explicitly bounding potential input profiles. Additionally, it embraces the concept of proportional input noise, a key characteristic of the stochastic optimized-submovement model.