The ability to notice erroneous behavior is crucial for effective training. Within the framework of neuroprosthetics, numerous studies in electroencephalography (EEG) confirm the existence of neural correlates when humans perceive the erroneous actions of the device. Subsequently, the decoding of this correlate has been used to correct the erroneous behavior performed by the agent or to tune the behavioral strategy of the agent, among others. However, a main limitation of current approaches is that the actions of the agent were discretized, thus restraining the usability of such systems. The main objective of this PhD study is to study, and decode, the neural correlates of error evaluation under continuous trajectories performed by external agents; and to use this decoding to tune the continuous behavior of the agent for individual users. To accomplish this goal, two essential questions will be investigated: (i) whether it is possible to infer individual preference under continuous state-action scenarios, and (ii) how to create a reliable decoding pipeline in a continuous fashion. Results obtained during the first year of the PhD have confirmed the existence of such correlates under continuous motions of a robotic arm. Furthermore, such correlates encode individual preferences, indicating that the neural prosthesis can be also customized for individual users, which may play an important rule to increase the quality of brain-computer based assistance. This property not only will increase the level of perceived assistance provided by a brain-computer interface, but also may facilitate embodiment of the brain-controlled device.