Optimization based Generation of
Walking Primitives and Step Sequences
for Perception Guided Biped Robots
This thesis presents a general method for the online generation of
step sequences and corresponding reference trajectories from path
information and environmental perception, which provide a basis for
stable and collision-free humanoid biped robot walking in 3D-scenarios
with prototypical obstacles. The reference trajectories are obtained
by a situation dependent online selection and concatenation of single
walking primitives, which are stored in an off/line computed
database. Optimal control techniques are employed for the systematic
synthesis of the walking primitives. Unilaterality
conditions between the feet and the ground, friction conditions,
restrictions of the joint drives as well as further restrictions of
the task space are taken into account by constraints to the
optimization problem. To allow a continuous locomotion of the walking
machine along a pre-specified local path and over obstacles like barrier,
ditch or stairs, a systematic graph based approach for situation
dependent online walking primitive selection is applied. Simulation
results and experiments with physical walking machines demonstrate the
practical relevance of the proposed concept.