A time-frequency analysis based on wavelet transform was carried out to extract error related local field potentials (LFPs) resulting from undesired action perturbations during the control of a robotic effector. During this study, a marmoset monkey controlled the movements of a robotic arm while periodically observing the arm unexpectedly moving to the incorrect target during perturbations. LFPs were recorded using a multi-electrode array implanted in the striatum. Results showed that the proposed signal processing method with a chosen classification threshold can effectively separate the perturbed and the normal situations. Obtaining such information during brain machine interfaces (BMI) tasks could improve the performance of BMI systems.