From Heat to Force: Retrospective Estimation of Finger Contact Forces from Thermal Signatures

Abstract

Contact forces are essential for understanding manipulation, tactile interaction, and robotic control. Measuring these forces, however, remains challenging due to the limitations of conventional sensors, which are often costly, geometry-dependent, and impractical in complex interactions. Furthermore, traditional force-sensing approaches, such as sensor gloves or sensorized objects, can bias interaction strategies, while existing non-invasive, optical approaches typically require detailed capture of the contact action. In this paper, we present a novel, non-invasive, non-instrumented approach for deriving contact forces from thermal imaging through retrospective force estimation without requiring direct capture of the contact event. We introduce an experimental setup and procedure for repeatable data acquisition, and provide an open data set comprising thermal contact patterns of finger pads recorded over various time periods, including precise force measurements up to 20N. We then demonstrate how the relationship between thermal signatures and contact properties — including force magnitude and contact duration — can be modeled. Our analysis reveals that thermal signatures encode sufficient information to reliably predict single-finger contact forces and durations, demonstrating that contact forces can be estimated up to 10 seconds after contact using a 500ms observation window, thereby enabling non-intrusive force estimation in human–object interaction.