This paper investigates the data-driven modelling of electrical power consumption in industrial robots performing different types of motion at various velocity levels. Three representative trajectory classes were analysed: joint motion, linear motion, and circular joint motion. Each class was executed at several commanded speeds. Instantaneous electrical power was measured during experimental runs, pre-processed, and used to identify parametric models relating power demand to selected motion-related variables. The identified model parameters are highly statistically significant, confirming the relevance of the adopted structure. Separate models calibrated for individual motion types achieved very good agreement with the empirical data. For joint and linear motion, correlation coefficients of approximately =0.95 and =0.93 were obtained, with mean absolute percentage errors below 2%. For circular motion, the fit remains satisfactory, with R ≈ 0.85 and MAPE below 4%. A unified model, identified using combined data from all motion types, can also reproduce the main trends in power consumption, even though with higher errors (MAPE above 5%), which illustrates the trade-off between accuracy and universality. These results show that relatively simple empirical models can accurately predict instantaneous power consumption. Such models could be employed for cost estimation, energy-oriented optimisation of robot trajectories, and implementing in industrial practice.