Real-Time Gait Planning and Control of Bipedal Mechanisms: Applications for a Transformable-Wheel Robot

In this thesis, we develop a novel method for generating real-time orbital trajectories for bipedal robots in dual-stance or single-stance. We then apply this method to a class of balancing robots with reconfigurable wheel profiles. The hybrid locomotion mechanism in question allows a wheel rim to unfold into a set of radially-projecting claws, which assists with obstacle traversal. Our method is based on some modest modeling assumptions, and exploits the Adomian decomposition method to determine an approximate solution to the robot dynamics. As a result, we derive a unique manifold that represents the set of Poincaré fixed points. This structure, which we term a fixed-point map (FPM) is parameterised by the gait period T, and describes the set of pendular initial conditions such that, after the completion of a step, the pendulum returns to the same conditions. To demonstrate the use of the FPM, we develop a model-predictive controller (MPC) which drives the robot state onto the FPM surface from large local neighborhoods. Controller performance is evaluated on a novel, high-fidelity simulation of a transformable-wheel robot in 3D space. Our controller was able to maintain balance of the robot during obstacle ascent and descent, while differentially steering the wheels, during dynamic, differential expansion of the wheels. These highly-complex motions were enacted under the influence of artificial sensor errors, external disturbances and parameter uncertainties.