Personalization Framework for Adaptive Robotic Feeding Assistance

Gerard Canal, Guillem Alenyà and Carme Torras

Abstract: The deployment of robots at home must involve robots with pre-defined skills and the capability of personalizing their behavior by nonexpert users. A framework to tackle this personalization is presented and applied to an automatic feeding task. The personalization involves the caregiver providing several examples of feeding using Learning-by-Demostration, and a ProMP formalism to compute an overall trajectory and the variance along the path. Experiments show the validity of the approach in generating different feeding motions to adapt to user's preferences automatically extracting the relevant task parameters. The importance of the nature of the demonstrations is also assessed, and two different training strategies are compared.

The FUTE framework overview

We present a three-phase framework, the “FUTE framework”, to design and develop adaptive assistive applications such as feeding or dressing a person. It comprises the following phases:

  1. Factory setting: the robot is provided with the skills needed to perform the assistive task in a generic way.
  2. User Tailoring: the behavior of the robot is adapted at the user's home in order to fulfil his/her specific needs.
  3. Execution tunning: the robot performs the task designed in the first phase but taking into account the personalization introduced in the second one, as well as the person's pose and other particularities of the task.

Demonstration: personalized assistive feeding

The framework was applied in an assistive feeding application.

Once a robot able to feed a person in a generic way (from the Factory setting phase) arrives at home, the caregiver must be able to adapt the robot behavior to the user needs or preferences (User Tailoring). For instance, a specific user, who is able to move his head, may prefer to fetch the food by himself rather than having it introduced in his mouth. So, as it can be seen in the following video, the caregiver is able to physically interact with the robot in order to modify the default robot movement so it ends outside of the mouth. This is done by stopping the robot execution until the user bites the spoon in the example.

Then, after repeating five times the adaptation, a new Probabilistic Movement Primitive (ProMP) that represents the new movement is learnt. With this, the user can now eat with the help of the robot in a more suitable way. In the next video, we can see how the robot would feed a user who is not able too move much once the personalization has been performed.