The RowBots are coming!
We have received a variety of reactions to the term and idea behind “RowBots”. These reactions tend to distribute along three lines: some like the witty wordplay, some quickly dismiss the concept, and some grasp the idea and ponder the potential. This blog post is for the latter group.
Rowers appreciate that rowing is simultaneously simple in concept but exceedingly complex in execution. It is simple in that the gross motions are pretty basic and relatively well-defined. Dedicated rowers will accumulate millions of strokes in their lifetime. But yet we can all still improve. Why? Because rowing stroke performance is highly affected by even tiny differences in execution. Exactly how you sequence your body movements, control your slide, or match up with others in the boat significantly effects how efficiently you move a boat. And therein lies the opportunity for computers, data analytics, artificial intelligence, bots – or the proclaimed “RowBot”.
Here is the high-level primer on machine learning, artificial intelligence, and bots. Supervised machine learning is a well-established method whereby a human expert reviews and “classifies” information or events. For example, a rowing coach might look at a rower lunge their body to the stern at the catch in an effort to add to the oar catch angle or to effectively store some energy that will be used in impending rebound for the drive. A coach would classify that as “lunging at the catch” and would focus the rower on developing a more controlled (“quieter”) upper body at the end of the recovery. Other commonly recognized coaching classifiers related to body and slide control include opening with the back, shooting the slide, opening too early/late with the back, yanking at the finish, too much/little layback, out of bow too fast/slow, rushing the slide. When these event “classifications” are correlated with data, patterns numerically emerge that can be identified by computer algorithms. The resulting model is the best replication of expert classification possible. The quality of the model depends on the quality of the data, the expert, and the chosen machine learning technique. And similar to statistics, the reliability of the model improves as it is fed more and better data.
We are currently building 2 different forms of RowBots: a general and a personal. The general RowBot is built on data from the rowing population at large. As such, it will be relatively basic in its diagnostic and coaching capabilities. The personal RowBot is unique to a specific individual and is built on the reports and coaching that they receive as paid SwingRow subscribers. The readiness and deployment of the RowBot necessarily depends on the quality of the data that is collected and the number of reports and coaching completed.
Yes, the RowBots are coming. Early adopters wanted!