Toward Shared Control in Clutter

Kurzfassung

Assistive robots can enable people with motor impairments to complete everyday tasks independently, offering active, semi-autonomous support. However, providing correct and seamless assistance in more cluttered multi-task environments remains a key open challenge. To address this challenge, we introduce a task-space-dependent clutteredness metric that quantifies scene complexity considering all tasks in the scene. This metric allows us to adapt shared control skills based on the scene's clutteredness with our new skill adaptation method ShaCC (Shared Control in Clutter). ShaCC comprises two steps: first, we optimize task space parameters to reduce task space overlap (Shrinking), and second, provide early orientation support toward the whole scene (Merging). We apply ShaCC to the Shared Control Template (SCT) framework for experimental evaluation. Our results show that the clutteredness metric is an expressive measure of scene complexity. Further, the task-aware skill representation of ShaCC improves assistance in multi-task environments with higher success rates and smoother task switching during manipulation. We show this in simulation with up to ten tasks per scene, improving the success rate by 45% on average, and validate ShaCC in a pilot study with three users on the assistive robot MAYA, achieving 100% task success in a real-world scene with five tasks.