Abstract: The following study presents the development of an algorithm of recognition, grip detection and trajectory planning for a robot of three degrees of freedom where objects are recognized by Convolutional Neural Networks (CNN) and gripping detection by geometric analysis of the object. The algorithm works on a non-controlled environment where it receives the images through a webcam, segments all the objects that are found in them, classifies them into one of three categories of tools (scalpel, scissors, screwdriver) trained on the CNN and searches for the tool desired by the user on which a feasible gripping point is selected and a path is executed that allows the manipulator to take the found object and move it to another point. Finally, functional tests are presented for the trained categories and the results are analyzed to determine grip accuracy in the real environment.