Is dressing a daily chore, in a range that includes cooking, cleaning, and eating, whose automation might have an enormous impact, especially because of aging? This is unclear, but nevertheless the progress in this task may be extrapolated to other tasks, or lead to niches of applications (for example, in more controlled industrial scenarios). Dressing another person is a difficult task even for humans; it requires coordination with the person being dressed (who can cooperate, remain passive, or even stiffen).

Surprisingly, the techniques in this paper are based on classical artificial intelligence (AI) approaches (hidden semi-Markov models (HSMMs), Viterbi, expectation-maximization (EM) algorithm). It’s comforting to see how far these techniques can reach, but the paper does not clarify what are the crucial bits that make them successful or better than previous approaches (or more data-intensive approaches, such as deep learning). The main contribution is said to be the augmenting of the movement primitives with sensory data.

The abstract seems to convey that this goes much beyond some controlled situations. But the experiments are about helping a user put on the sleeve of a jacket (with two positions of the hand) and put on a shoe (with possible obstacles). There lies the question of how realistic this is, as the system ignores the physics of clothes (slack, fabrics, elasticity, buttons, and so on) and bodies (joints, weights, and so on). The main remaining question is whether this approach will be scalable to more realistic scenarios. Will more sophisticated learning and cognition be needed in the end?