Neurorobotic Technology for Advanced Robot Motor Control

Legged robots have the potential to go anywhere that their biological counterparts can go and traverse terrain inaccessible by most wheeled robots. They can interact with harsh terrain with many obstacles or generic physical environments designed for legged locomotion. This is because legged robots are equipped with multiple legs that each has multiple degrees of freedom. While this redundancy creates high adaptability and robustness to failures, it also complicates the controller design. There are, in general, three fundamental adaptation types to be considered when designing a locomotion controller; frequency, phase, and amplitude adaptation. The hypothesis is that a central pattern generator (CPG) based locomotion controller capable of all three types of adaptations will enable 1) savings in energy, 2) a general and comprehensible controller architecture, 3) adaptive locomotion, and 4) emergent complex behaviors.


This thesis consists of five papers in which locomotion control mechanisms implementing the three fundamental types of adaptation are presented. Together, the mechanisms fill an important gap in the research field of locomotion control by being simple yet capable of generating complex locomotion behaviors.
Moreover, the mechanisms are expandable, which makes it possible to extend them with new behaviors. This is especially useful in a fast­growing field like locomotion control, where new advances are introduced all the time.


The first four papers develop adaptive mechanisms for frequency and motor pattern adaptation. Both mechanisms are based on a bio­inspired artificial CPG. The adaptive frequency mechanism can adapt the CPG frequency within seconds based on the robot body as well as external and internal perturbations. Doing so enables a robot to move energy­efficiently, prevent damage, and minimize tracking error. The adaptive motor pattern mechanism can facilitate both phase and amplitude adaptation. It can learn motor
patterns within minutes based on the robot body, external perturbations, and the desired behavior. Furthermore, the mechanism is modular, meaning that it can be expanded with additional functionality based on the mission at hand. Being modular also allows for the removal of modules that relies on broken or faulty sensors. The final paper of the thesis is on the combination of the frequency and motor pattern adaptation mechanisms. Combining them is straightforward, as both mechanisms are using a CPG as a core component. Results show that the mechanisms complement each other, and their combination can be seen as an essential step for further studies on adaptive locomotion control.