Smooth Real-Time Walking-Pattern Generation for Humanoid Robot LOLA

This video demonstrates our new approach for planning smooth center-of-mass trajectories for biped walking robots. The method is based on quintic spline interpolation and collocation and generates dynamically and kinematically feasible motions in real-time. Method/Testscenario: Our humanoid robot Lola steps up and down a platform of 12.5cm height. The planned center-of-mass motion respects the dynamics and kinematic limits of the robot using simplified models. The complete motion lasts more than 17 seconds and is planned in less than 9 milliseconds (CPU only, single-core). Simulation: Custom multi-body simulation Visualization with Blender (custom interface using Blenders Python API) Experiments: Planning and control of the robot runs in real-time (onboard). External communication only triggers a signal to start and stop walking. The vision system is not active, thus foothold sequence is predefined. This work is supported by the German Research Foundation (DFG, project number 407378162).