Optimal strategies for kiiking: active pumping to invert a swing

TL;DR

This paper models kiiking as an actively controlled pendulum, deriving optimal pumping strategies using control theory and reinforcement learning, and validates the model with experimental data considering air drag effects.

Contribution

It introduces a minimal control model for kiiking, linking optimal control and reinforcement learning to determine effective pumping strategies.

Findings

Optimal control strategies resemble greedy algorithms maximizing energy gain.

Reinforcement learning aligns with theoretical optimal strategies.

Model predictions match experimental data when air drag is included.

Abstract

Kiiking is an extreme sport in which athletes alternate between standing and squatting to pump a standing swing till it is inverted and completes a rotation. A minimal model of the sport may be cast in terms of the control of an actively driven pendulum of varying length to determine optimal strategies. We show that an optimal control perspective, subject to known biological constraints, yields time-optimal control strategy similar to a greedy algorithm that aims to maximize the potential energy gain at the end of every cycle. A reinforcement learning algorithms with a simple reward is consistent with the optimal control strategy. When accounting for air drag, our theoretical framework is quantitatively consistent with experimental observations while pointing to the ultimate limits of kiiking performance.