Handling materials on construction sites is one of the most critical aspects of a construction project. In low-rise buildings and early-stage construction sites without lifts or material hoists, transfer of materials heavily relies on manual operations. Proper material handling is imperative in ensuring the safety and productivity of workers. To identify whether the workers are handling materials properly, a foot plantar sensor can be employed to measure the workers’ workload and estimate their working posture. Despite vast research that adopted foot plantar sensors to monitor the workers’ workload when handling construction materials, further study is needed to understand the workload of carrying construction material on staircases. This study investigates whether workload monitoring methods can be applied to the material-carrying works on staircases. Foot plantar pressure data of workers with four postures is monitored and collected. The recurrent neural network (RNN) based deep learning network, which is gate recurrent unit (GRU) model, is adopted to estimate worker’s postures using foot plantar pressure data. Results demonstrated that this methodology could train an estimation model to predict worker’s postures for handling materials on staircases. The average training time for training the estimation model is 18 min. The model’s average accuracy rates in the training, validation, and testing sets are 80%, 88%, and 80% respectively. The results revealed that GRU could train estimation models with limited plantar pressure data in a fast and effective manner. The study confirmed that the foot pressure value can be derived subject to worker postures when handling construction materials on staircases.