The video above demonstrates a cool new automatic data-driven ghosting system for plotting sports strategy, in this case soccer. This isn’t “ghosting” in the sense of social disappearance, but rather the use of translucent markers—i.e., ghosts—to envision likely gameplay outcomes. It’s a big step forward from a manual ghosting system developed in 2013 by the Toronto Raptors thanks to its advanced neural network.
The new system, developed by DisneyResearch working together with CalTech and STATS, offers coaches the chance to easily and quickly play out various “what if” scenarios. It shows the positions of actual players in previous games, along with ghosts created by what DisneyResearch calls “deep imitation learning” that suggest what some other team might have done. Is that a statistically generated average team, a fantasy team, or individual players from a real one? Coach’s choice.