Using bilateral symmetry of the biped robot mechanism for efficient and faster optimal gait learning on uneven terrain

This paper, describes the relation between dynamically similar right gait (RG) and left gait (LG) of a 14 degrees of freedom (DOF) biped robot on uneven terrain by using the dynamic similarity principle of gaits (DSPG). The existence of DSPG is ingrained in bipedal systems’ bilateral body plan and the mirror image transformation of polar and axial vectors. The concept of fundamental basis vectors associated with the support foot plane act as a reference frame to compare the dynamic similarity among gaits for different walking conditions on uneven terrain. The DSPG discerns the generation of all possible optimal gaits on uneven terrain from a small primitive set of gaits. The identification of primitive walk dataset results in saving of both walk dataset generation time and learning time. The proposed DSPG is validated by generating the optimal RG of a 14-DOF biped robot using a genetic algorithm (GA) and utilizing DSPG to obtain corresponding dynamically similar LG. Simulation experiments of the bipedal walk on uneven terrain is validated by generating a primitive walk dataset for a 14-DOF biped robot and learning it by a feed forward neural-network (NN). Using the trained NN and DSPG, biped robot can successfully walk on 3D uneven ground by generating real-time optimal gaits.