State-of-the-art wearable lower-limb robot controllers typically use established baseline human kinematics during common mobility tasks. Unfortunately due to the variability in human response during perturbations, these lower-limb controllers are unable to effectively assist with perturbation recovery. Accurate and quick predictions of kine-matic responses to unexpected disturbances during motion can help assistive robotic devices safely aid with an individual's recovery. This paper presents three methods for predicting swing hip kinematics during trip recovery: a Gaussian process regression (GPR) model; a time-series neural network; and a pendulum model with linear feedback. Data were collected in an experiment where 16 subjects were tripped at random percentages of swing phase. The three prediction methods were applied to these data and evaluated for simulation accuracy and computation time. Both subject-specific and generalized models were investigated. Results indicate that the GPR model is the best choice for kinematic predictions due to its low simulation error in both subject-specific and generalized cases and lowest computation time.