Version 1
|
Pattern wristband worn by the user
|
Neural networks interacting with a hard-coded taxonomy
|
Audio prompts, with one prompt for each step
|
10
|
[22, 33]
|
Version 2
|
Using background subtraction to isolate the user's hands. Tracking of hands and task objects (i.e. soap and towel) using a preset colour model.
|
Fully observable Markov decision process (MDP). This technique assumes the world is fully observable; it does not take into account hidden variables, such as user responsiveness.
|
Audio prompts with three levels of assistance (minimal, moderate, and maximal) for each step.
|
4
|
[20, 37–39]
|
Version 3 (system presented in this paper)
|
Colour based flocking.
|
[Belief monitoring system & policy] Partially observable Markov decision process (POMDP). This model takes into account hidden variables and is able to make decisions in conditions of uncertainty.
|
Audio and audio-video prompts with three levels of assistance (minimal, maximal, and maximal + video demonstration) for each step. Encouragement and a reminder regarding the activity the user is attempting added. Professional actor recorded prompts.
|
6
|
[18, 19, 21, 23, 25]
|