Title: Car racing strategy
Authors: Stavros Tsalidis - Quantumblack Mckinsey (United Kingdom) [presenting]
Abstract: Car racing is a popular competition in which fully or partially electric powered cars race a fixed period of time and/or race laps. The total energy allowed to be used in the race is fixed or restricted and has to be managed wisely in order to win points by overtaking but not run out of energy before end of race. A racing strategy consists in allocating energy budgets dynamically at each lap start for optimal positioning. We build an environment simulating the effects of the energy allocation on the state of a race and use simulations to investigate the outcome of strategies and scenarios in energy budgeting. Parameters of simulations for different drivers are estimated from past races data. We apply reinforcement learning techniques to estimate optimal policies/strategies for energy allocation. The simulation environment is used to validate the estimated optimal strategies.