The main goal of this thesis was to create a system that enables a humanoid robot to imitate human walking (chapter four), which is captured by an RGB-D camera like for example Microsoft Kinect. We also considered how to prevent collisions between the limbs of a humanoid robot during imitation.
In the second chapter we discuss the main components that are used in our work. These are humanoid robot HOAP-3 and Microsoft Kinect.
Humanoid robot is a robot whose structure is similar to the structure of a human body. Humanoid robots are typically comprised of the torso and the head, two arms, and two legs. Among the most important elements that define the capabilities of humanoid robots are: 1) sensors and actuators used to build the robot, 2) ability to provide energy for autonomous operation, 3) ability to effectively learn complex tasks with or without external teacher intervention, including movement adaptation to the enviroment, 4) ability to securely operate in human environments, without compromising safety of people in the area.
Fujitsu HOAP-3 is a humanoid robot distinguished by its compactness, low weight, and easy operation. It has 28 joints with electric servomotors, two cameras to provide visual information, microphone, three axis gyroscope, accelerometer, force sensors on the soles of the feet, distance meter, and a built-in module for recognition and speech synthesis.
Microsoft Kinect is an RGB-D camera that can be applied to detect and follow human motion in 3-D space. Range images are essential to enable Kinect to track up to six people and up to 25 joints per person. In our work we used Kinect to capture the motion of a human demonstrator. In the third chapter follows the human hands movement imitation, preventing collisions between robot segments and descritption of stability control.
Due to its similarity to people, a humanoid robot can imitate human motion captured by Kinect. For a more effective movement imitation, two additional problems need to be solved: collision avoidance and stability of the robot. In this thesis collision avoidance was implemented using the task priority approach. In this context, the primary task is to prevent collisions between the robot body parts, The secondary task is to imitate the movement acquired by Kinect.
For effective imitation of human walking, the robot must remain stable while imitating the demonstrator's motion. The robot is most stable when the zero moment point (ZMP) is in the center of the support polygon that surrounds the legs of the robot. Thus when imitating human body motion while walking, the primary task must ensure the stability of the robot, while the secondary task refers to the imitation of the demonstrator's movement. ZMP depends on the center of gravity of the robot. All this is important for planning walking trajectories that are used in our work in chapter four.