 This paper proposes two new deep learning, DL, techniques for decoding EMG signals for use in MoMAs. These techniques, temporal multi-channel transformers and vision transformers were compared against traditional random forest, RF, models and convolutional neural networks, CNNs, which are commonly used in MoMAs. The proposed DL techniques outperformed both RF and CNN models in terms of accuracy and speed of decoding the motion. Additionally, the proposed DL techniques were able to generalize across different motion-object combinations, while the other models could not. This article was authored by Ricardo V. Godoy, Ananita Vetti, and mine is Liero copies.