Control of Adaptive Locomotion for Quadruped Robot CENTAURO

When deployed in the real world, quadruped robots are expected to adapt to various working conditions so as to accomplish the targeted missions successfully. In this thesis, we aim to study the problems of adaptive legged locomotion and develop control schemes that will enable our quadruped robot CENTAURO to perform adaptation in different tasks and scenarios. To be specific, three kinds of adaptation for quadruped robots are studied in this thesis and will be introduced as follows. The adaptation to unknown payloads during locomotion is first considered. To begin with, a position-based control pipeline for quasi-static legged locomotion is first developed, consisting of a Cartesian motion planner and a hierarchical IK controller. Two types of disturbance observers are implemented to estimate the disturbances induced by the payloads and are then compared in simulation tests. An integrated scheme is finally validated on CENTAURO, demonstrating that the robot manages to adapt to 20~kg payloads and walk stably in different scenarios. The second case considered is to traverse uneven terrains without perception aided. To this end, the Cartesian motion planner is extended by including contact-triggered touchdown adaptation that depends only on proprioceptive information. The updated control pipeline is validated in experiments where the robot is able to walk up a ramp without any knowledge of the terrain geometries beforehand. The third case of adaptation concerns online modulation of joint stiffness during locomotion, which is a kind of bio-inspired adaptation that enables compliant while robust interactions with surroundings. An intuitive modulation algorithm that depends on predefined reference stiffness and tracking errors is first proposed, the improvement of which is demonstrated in experiments but theoretical criteria such as optimality or stability are not considered. A novel scheme is then proposed to achieve optimal joint stiffness modulation, where the stability issue of variable stiffness control is theoretically resolved by using an energy tank-based constraint. The integrated control pipeline is finally validated on CENTAURO, demonstrating that the robot achieves a balance between robust motion tracking and interaction compliance. Intensive experimental tests are carried out to validate the effectiveness of the developed schemes on hardware. Thus the results presented in this thesis should eventually push forward the quadruped robot CENTAURO towards applications in the real world with required capabilities of adaptation.