The European Project MUltimodal Neuroprosthesis for Daily Upper limb Support (MUNDUS) aims at the development of an assistive platform for recovering direct interaction capability during daily life activities based on arm reaching and hand functions. Within this project the present study is focused on the design of a biomimetic controller able to modulate the neuromuscular electrical stimulation needed to perform reaching movements supported by a commercial passive exoskeleton for weight relief. Once defined the activities of daily life to be supported by the MUNDUS system, an experimental campaign on healthy subjects was carried out to identify the repeatable kinematics and muscular solution adopted during the target movements. The kinematics resulted to be highly stereotyped, a root mean squared error lower than 5° was found between all the trajectories obtained by healthy subjects in the same movement. A principal component analysis was performed on the EMG signals: less than 5 components explained more than the 85% of the signal variance. This result suggested that the muscular strategy adopted by healthy subjects was stereotyped and can be replicated by a biomimetic NMES controller. The controller was based on a time-delay artificial neural network which mapped the dynamic and non-linear relationship between kinematics and EMG activations to determine the stimulation timing. The stimulation levels reproduced the same scaling factors found between muscles in the stereotyped strategy. The controller was tested on 2 healthy subjects and though it was a feedforward controller, it showed good accuracy in reaching the desired target positions. The integration of a feedback controller is foreseen to ensure the complete accomplishment of the task and to compensate for unpredictable conditions such as muscular fatigue.