Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Bio-inspired motion control of the humanoid robot NAO using Central Pattern Generators (CPGs)

Kousanakis Vasileios

Full record

---

Summary

Recent developments in computational and hardware systems enabled the development of more complex artificial systems. In particular, this has brought to life new challenges in the field of robotics, such as the development of humanoid robots with the ability to interact with complex environments, like those where humans live and act. Research has addressed the locomotion of such robots and the way they interact with humans. However, the increasing complexity in the design of the body of humanoid robots, necessitates the development of more satisfactory control methodologies than the existing ones. For that purpose, research in robotics has attempted to get inspired by the study of living organisms, which demonstrate robust and adaptable locomotor behavior. Key components of their motion control are neural networks, called Central Pattern Generators (CPGs), such as those located on the spine of the vertebrate bodies, and are responsible for the production of rhythmic control signals during walking, running, swimming or flying, even in the absence of sensory feedback. Reproducing robustly such motion control schemes in humanoid robots represents a significant challenge, made even more important from the requirement of controlling fast and efficiently a large number of degrees of freedom in such robots. In this thesis, we examined through robotic experiments on the NAO humanoid robot, the robustness of its built-in walking behavior on steps, inclines, as well as on rugged and granular substrates. The robot and its built-in walking behavior are developed by the company Aldebaran Robotics. The analysis of the experiments showed that the robot has some difficulty in dealing with such substrates. At this point, the idea of using a completely different approach to create a walking behavior for the robot, came up. An attempt is made to set up and exploit a CPGs-based motion control scheme, as an alternative way of achieving a stable walking locomotion for the simulated humanoid robot NAO, using the Webots simulator. The type of Central Pattern Generators selected is based on the Hopf nonlinear oscillator, which has properties useful for the control of the joints of the robot during walking, and is able to reproduce them in a satisfactory way. The procedure followed starts with the recording of the joint trajectories of the robot during walking, then these are learned by the CPGs via a Hebbian learning process, and, then, the CPGs provide desired joint trajectories to the robot, in order to reproduce the walking gait. Furthermore, an optimization process based on genetic algorithms is employed, to achieve the fine tuning of the CPGs via a well-defined objective function.

Available Files

Services