Data-Driven Kinaesthetic Action Classification for Remote Teleoperation over the Tactile Internet

Abstract:
This paper presents a set of heuristic methods for real-time action classification based purely on physics data from a kinaesthetic interaction, aimed at improving interaction understanding in haptic-enabled systems. Three individual heuristics— Statistically-Based Mean and Deviation (SBMD), Axis-Aligned Rule-Based (AARB), and Transient Peak-Based Interaction Detection (TPBID)—analyze velocity and force signals over a fixed time window using statistical patterns, directional trends, and peak detection strategies. Each heuristic demonstrates high performance for specific interaction types such as tapping, dragging, or pressing. To enhance overall reliability, a hybrid approach is proposed, combining the outputs of the three individual heuristics to produce a unified classification. Experimental validation with 21 users (aged 20-33 years) performing five distinct actions shows that the hybrid method achieves over 96% overall accuracy. These results highlight the potential of the proposed framework for integration in stability control and adaptive network resource allocation in future communication systems for the Tactile Internet.