Abstract:
The connectionist central pattern generator (CCPG) model is suitable for controlling robots and generating gaits, however, the traditional CCPGs can't generate the 3D gaits well. To solve this problem, an improved neuron model and an improved hierarchical CCPG (HCCPG) model are proposed according to biology principles. HCCPG can generate the phase-coordinated multi-degrees-of-freedom motion control signals well, so it solves the gait generation problem in traditional CCPGs. Based on the HCCPG, a unified generation method is proposed for 2D gaits and 3D gaits. The properties of turning gait are investigated systematically and thoroughly to make better use of it to adapt to narrow curved passages. The proposed HCCPG model and the derived gait properties are useful for improving the robot's adaptability.