Optimal Learning for Rowing: Minimizing Race Time

Abstract

Athletes across many different sport disciplines, as well as anyone who has participated in a race long enough to feel fatigued, are all faced with the challenge of pacing themselves in order to achieve their fastest possible time. Going out too fast can lead to excess lactic acid buildup that slows you down later on, while going out too slow risks never reaching your maximum potential speed. This thesis examines an individual athlete’s decision of how to expend themselves over the course of a race in order to achieve their optimal time. Specifically, this paper focuses on determining the optimal policy to minimize time to complete a 2000 meter (m) rowing race. The model assumes that each individual athlete has a unique physiology and therefore a unique optimal pacing strategy based on how their body responds to exercise. By modeling this problem as a stochastic decision problem, we learn how an athlete’s lactic acid levels, heart rate, and adrenaline function over the course of a race and use this information to determine their optimal pacing. The pacing dynamically updates after each 500m race increment has been completed. Applying this mathematical framework to athletics, where pacing is usually considered a skill of an experienced athlete, can improve performance at all levels of competition.