Intentional Target Reaching With an Underactuated Exoskeleton in a Cluttered Environment

Upper-limb exoskeletons can assist individuals with arm weakness, but delivering appropriate assistance remains challenging. Effective assistance requires accurately understanding the user's intent. We propose an Intentional Target (IT) estimation framework that integrates shear force-based forearm rotation detection, vision-based object recognition, and hand tracking. This approach enables precise IT identification in cluttered environments, specifically tabletops with multiple objects, resembling real-world daily activities. After identifying the IT, the required joint angles for reaching are estimated using a fusion model that combines the Human-Robot-Objects (HRO) model with human behavior data. These estimated angles serve as benchmarks for controlling portable underactuated exoskeletons, ensuring effective assistance while maintaining a compact design suitable for daily or home use. In trials with healthy adult participants, the system achieved a task success rate (TSR) of 92%, with certain objects, such as a cup, reaching 100% accuracy. These results validate the effectiveness of our IT estimation framework in cluttered environments and highlight the capability of our control strategy to overcome underactuation limitations, ensuring reliable reaching assistance.