Hardware, software and control design considerations towards low-cost compliant quadruped robots

Quadrupedal robots have been a field of interest the last few years, with many new maturing platforms. Many of these projects have in common the use of state of the art actuation and sensing, and therefore are able to handle difficult locomotion tasks very effectively. This work focuses on another trend of low-cost, quadrupedal robots, that features less precise actuators and sensors, but overcomes their limitations with strong bio-inspired designs to achieve state of the art locomotion. We aim here to further extend the achievements of this approach to handle more complex tasks and that require anticipation, We would like also to verify to which extent a close synergy between clever mechanics, sensorimotor coordination, and Central Pattern Generator models is able to handle these tasks. This thesis presents supporting work that was required to pursue this goal. A software architecture for the development of real-time drivers and low-level control for robotic applications, based on a clear separation of concerns is presented. An implementation of this architecture able to handle the specific requirements for small compliant quadruped robots is proposed. Furthermore, the development and integration of a communication protocol for inter-electronic devices communication on the Oncilla robot is discussed. As leg load is a key quantity in some of the sensory-motor coordination this thesis want to explore, a novel tactile sensing approach for its estimation is proposed, based on an Extended Kalman Filter data fusion of static and dynamic tactile sensor information. Then, to support the design of efficient interactions between the control and the bio-inspired mechanics, accurate dynamic modeling of the Advanced Spring Loaded Pantographic leg, equipping all robots considered here, is presented. We propose two approaches to this modeling with the presentation of their benefits and limitations. Finally, two Central Pattern Generator architectures are proposed, based on biologically inspired foot trajectories. The first is using a well-known method for inter-limb coordination with strong neural coupling, and the second, the Tegotae rule, relies only on limb physical coupling and strong sensory-motor coordination. These two approaches are compared on their capacity to handle dynamic footstep placement and it let to the conclusion that strong sensory-motor coordination is required for this task.