In this study, seven able-bodied human subjects controlled a robotic gripper with surface electromyography (sEMG) activity from the biceps. While subjects controlled the gripper, they felt the forces measured by the robotic gripper through an exoskeleton fitted on their non-dominant left arm. Subjects were instructed to identify objects with the force feedback provided by the exoskeleton. While subjects operated the robotic gripper, scalp electroencephalography (EEG) and functional near infrared spectroscopy (fNIRS) were recorded. We developed neural decoders that used scalp EEG to reconstruct the sEMG used to control the robotic gripper. The neural decoders used a genetic algorithm embedded in a linear model with memory to reconstruct the sEMG from a plurality of EEG channels. The performance of the decoders, measured with Pearson correlation coefficients (median r-value = 0.59, maximum r-value = 0.91) was found to be comparable to previous studies that reconstructed sEMG linear envelopes from neural activity recorded with invasive techniques. These results show the feasibility of developing EEG-based neural interfaces that in turn could be used to control a robotic device.