Stick balancing and human upright stance are complex motor tasks that require the central nervous system (CNS) to stabilize unstable dynamics of controlled objects. For those tasks, the CNS must utilize a well-designed control strategy to compensate multiple factors of instability. It has been reported that a type of intermittent time-delayed feedback control can avoid delay-induced instability and reproduce long-term correlations in stochastic postural sway patterns as well as non-Gaussianity in the acceleration of sway. A key mechanism of those superior capabilities of the intermittent control model is to exploit transient contracting dynamics near a stable manifold of the unstable upright equilibrium, which becomes available for the intermittent controller by switching off the feedback control in a state-dependent manner. Here, to characterize human control strategies during stick balancing, we developed an experimental system of a visuo-motor task, in which participants control a virtual stick using a joystick. We then established a framework for assimilating the intermittent control model into sway time series data acquired from each participant based on the Sequential Monte Carlo (SMC) approximate Bayesian computation (ABC) for parameter inference of the model. As a result of SMC-ABC, we showed that sway data from majority of participants were better fitted by the intermittent control with small feedback gains, whereas the rest by continuous control with no selections for switching the feedback controller off in the intermittent control model. Our results suggest that the intermittent control is a major strategy for the motor tasks of stabilizing unstable dynamics.