Analytic Bipedal Walking with Fused Angles and Corrective Actions in the Tilt Phase Space

Abstract

As humanoid robots start to move from research labs to workplace environments and homes, the topic of how they should best and most reliably locomote in the face of unknown disturbances will be a topic of increasing importance. This thesis presents algorithms and methods for the feedback-stabilised walking of bipedal humanoid robotic platforms, along with the underlying theoretical and sensorimotor frameworks required to achieve it. Bipedal walking is inherently complex and difficult to control due to the high level of nonlinearity and significant number of degrees of freedom of the concerned robots, the limited observability and controllability of the corresponding states, and—especially for low-cost robots—the combination of imperfect actuation with less-than-ideal sensing. <br /> The methods presented in this thesis deal with these issues in a multitude of ways, ranging from the development of an actuator control and feed-forward compensation scheme, to the implementation of numerous sensor calibration and processing schemes, to the inclusion of inherent filtering in almost all of the gait stabilisation feedback pipelines. Two gaits are developed and investigated in this work, the direct fused angle feedback gait, and the tilt phase controller. Both gaits follow the design philosophy of internally leveraging a semi-stable open-loop gait generator, and extending it through stabilising feedback via the means of so-called corrective actions. The idea of using corrective actions is to modify the generation of the open-loop joint waveforms in such a way that the balance of the robot is influenced and thereby ameliorated. Examples of such corrective actions include modifications of the arm swing and leg swing trajectories, the application of dynamic positional and rotational offsets to the hips and feet, and adjustments of the commanded step size and timing. <br /> Underpinning both feedback gaits and their corresponding gait generators are significant advances in the field of 3D rotation theory. These advances in particular include the development of three novel rotation representations, the tilt angles, fused angles, and tilt phase space representations. All three of these representations are founded on a new innovative way of splitting 3D rotations into their respective yaw and tilt components. <br /> All of the algorithms presented in this thesis were implemented as part of the Humanoid Open Platform ROS Software release, and tested on a multitude of real and simulated robots, including in particular the igus Humanoid Open Platform and NimbRo-OP2. The notable walking stability that was achieved critically contributed to Team NimbRo’s yearly wins at the international RoboCup competition from 2016 onwards.