Pion production is a significant component of the signal in accelerator-based neutrino experiments. Over the last years, the MiniBooNE, T2K and MINERvA collaborations have reported a substantial amount of data on (anti)neutrino-induced pion production on the nucleus. However, a comprehensive and consistent description of the whole data set is still missing. We aim at improving the current understanding of neutrino-induced pion production on the nucleus. To this end, the comparison of experimental data with theoretical predictions, preferably based on microscopic models, is essential to disentangle the different reaction mechanisms involved in the process. To describe single-pion production (SPP) we use a hybrid model that combines a low- and a high-energy approach. The low-energy model (LEM) contains resonances and background terms. At high invariant masses, a high-energy model based on a Regge approach is employed. The model is implemented in the nucleus using the relativistic plane wave impulse approximation (RPWIA). We present a comparison of the hybrid-RPWIA and LEM with the recent neutrino-induced charged current $1\pi^+$ production cross section on water reported by T2K. In order to judge the impact of final-state interactions (FSI) we confront our results with those of the NuWro Monte Carlo generator. The hybrid-RPWIA model and NuWro compare favorably to the data, albeit that FSI are not included in the former. These results complement our previous work [Phys. Rev. D 97, 013004 (2018)] where we compared the models to the MINERvA and MiniBooNE $1\pi^+$ data. The hybrid-RPWIA model tends to overpredict both the T2K and MINERvA data in kinematic regions where the largest suppression due to FSI is expected, and agrees remarkably well with the data in other kinematic regions. On the contrary, the MiniBooNE data is underpredicted over the whole kinematic range.