This paper presents the implementation of a simulation of a robotic arm whose task is to collect different objects in a virtual environment. To develop this task, the control of the robotic arm is done through 10 different hand gestures, which are recognized by a CNN with a structure type DAG Network (or DAG-CNN), reaching an accuracy of 84.5% in the recognition of gestures. Likewise, real-time tests are carried out on the already trained network, where the user is in a semicontrolled environment indicating the different actions for the robot to perform, where the correct operation of the trained network was verified, obtaining a high precision in the recognition of the commands made, that is, without errors in the control actions followed by the robot.

No abstract

Abstract-This article presents a supervised control position system, based on image processing and oriented to the cooperative work between two humanoid robots that work autonomously. The first robot picks up an object, carry it to the second robot and after that the same second robot places it in an endpoint, this is achieved through doing movements in straight line trajectories and turns of 180 degrees. Using for this the Microsoft Kinect , finding for each robot and the reference object its exact spatial position, through the color space conversion and filtering, derived from the information of the RGB camera that counts and obtains this result using the information transmitted from the depth sensor, obtaining the final location of each. Through programming in C #, and the developed algorithms that allow to command each robot in order to work together for transport the reference object, from an initial point, delivering this object from one robot to the other and depositing it in an endpoint. This experiment was tested performed the same trajectory, under uniform light conditions, achieving each time the successful delivering of the object.

No abstract

No abstract