Investigating the Levels of Autonomy for Personalization in Assistive Robotics

摘要: Without a proper balance between robot autonomy and user control, assistive robots can be over-engineered or put too much burden on users. Determining ideal levels of autonomy is important when personalizing assistive robots to individual users. For example, contemporary robot-assisted feeding (RAF) systems encounter challenges when users sit in positions other than directly facing a table if this was its original design requirement. Solutions like manual fixation of sitting or plate positions temporarily work, but these may not sustainably ensure effectiveness and user comfort. Prior work has studied levels of autonomy desired in RAF systems but has focused on functional tasks such as food transfer. In this work, we begin investigating whether these findings hold during personalization algorithms, which may need to change outcomes and procedures based on the user and location. We tested out levels of desired autonomy in an example personalization learning algorithm with one potential user of an RAF system on the test case of adjusting to different sitting positions. Our preliminary results in this case study suggest that shared autonomy can be a good starting point for personalizing RAF systems.