Human-Like Trajectories Generation via Receding Horizon Tracking Applied to the TickTacking Interface

TL;DR

This work addresses generating human-like trajectories for a rhythm-based TickTacking interface used in a target-tracking task. It develops a receding horizon control framework that injects human-inspired features—preferred diagonal directions, magnitude constraints via the $l_1$ norm, and control sparsity—into a quadratic program solved at each step. The human-like RH controller produces trajectories that resemble human data and can outperform a baseline optimal controller in aligning motion statistics while maintaining tracking performance. The approach offers design insights for intuitive rhythm-based HCI and training tools, with potential applicability to broader rhythmic interfaces and user-guided adaptation.

Abstract

TickTacking is a rhythm-based interface that allows users to control a pointer in a two-dimensional space through dual-button tapping. This paper investigates the generation of human-like trajectories using a receding horizon approach applied to the TickTacking interface in a target-tracking task. By analyzing user-generated trajectories, we identify key human behavioral features and incorporate them in a controller that mimics these behaviors. The performance of this human-inspired controller is evaluated against a baseline optimal-control-based agent, demonstrating the importance of specific control features for achieving human-like interaction. These findings contribute to the broader goal of developing rhythm-based human-machine interfaces by offering design insights that enhance user performance, improve intuitiveness, and reduce interaction frustration

Figures (6)

• Figure 1: Encoding of controlled velocity directions, in case duplets are used by the TickTacking interface.
• Figure 2: Histogram of velocity directions for the free navigation and target tracking scenarios in rocchesso2024spacetime, compared to the velocity directions of the reference trajectory. The bar height represents the relative frequency of observations.
• Figure 3: Features of the human-generated trajectories acquired in rocchesso2024spacetime (in addition to those shown in Fig. \ref{['fig:human_total_directions']}).
• Figure 4: Example trajectories generated by the baseline and human-inspired RH controllers. The controller and reference trajectories practically overlap in the baseline scenario. One example of human-generated trajectories is included for comparison. A dynamic visualization of these trajectories is available at https://doi.org/10.5281/zenodo.15240338
• Figure 5: Histogram of velocity directions in the baseline and human-like scenarios, compared to the velocity directions in the human tracking scenario and the reference trajectory. The velocity directions of the reference trajectory and the baseline scenario overlap. Bars corresponding to the cases are partially transparent to enhance the visibility of the other two cases. The bar height represents the relative number of observations.