Self-supervised learning (SSL) is relatively new learning paradigm in machine learning and robotics. You can have a look at this post https://ai.stackexchange.com/q/10623/2444 for more info.
I believe that SSL will be useful especially in cases where there is data coming from multiple different sources (e.g. different sensors). I think it will be particularly useful when you need to compute the likelihood of the data coming from a single data source (e.g. one sensor) by using the other available sources as “confirmatory” sources.
I think that there is a relation between SSL and HTM, but I would like to hear from the HTM experts what you think about it. Is HTM doing some kind of SSL? Given my knowledge of the HTM theory, I really think so. For example, in the paper that describes the example of the finger touching a cup. If you use multiple fingers (i.e. sensors) we more rapidly arrive at a conclusion regarding the object than by just using one finger.
I am also looking for possible applications of the HTM to robotics, in the context of SSL. I am open to suggestions.