Modeling the "Gorilla Arm" Effect with Reinforcement Learning

The “Gorilla arm” effect is a common problem of mid-air interaction which appears during excessive arm fatigue. To predict and prevent such problems at a low cost, we investigate user testing without real users, utilizing biomechanically simulated AI agents trained with Reinforcement Learning using a cumulative fatigue function from biomechanical literature. We show that the simulated fatigue data matches human perceived fatigue ratings based on the Borg CR10 scale. Our work demonstrates that deep RL combined with the fatigue model provides a viable tool for predicting both interaction movements and user experience in silico, without users. ACM Reference Format: Noshaba Cheema, Laura A. Frey-Law, Kourosh Naderi, Jaakko Lehtinen, Philipp Slusallek, and Perttu Hämäläinen. 2021. Modeling the "Gorilla Arm" Effect with Reinforcement Learning. In MIG’21: Motion, Interaction and Games. ACM, New York, NY, USA, 3 pages. https://doi.org/10.1145/1122445.