Abstract:
Legged locomotion holds numerous advantage over wheeled one on rough terrain. But computational complexity and use of fix pattern of leg movement in classical control method severely reduces the advantage of legged locomotion. Recent research suggests that a movement is composed of few simpler sub-movements called movement primitives. Dynamic Movement Primitives, DMP is one of the many way of introducing the concept of movement primitive to robotics.
Locomotion requires strong co-ordination within and between the legs, as well as continuous modulation of the trajectories. In this context, Dynamic Movement Primitive (DMP) provides one of the most powerful tool to represent and modulate a movement. In addition to that, different DMPs can be coupled together to achieve complex co-ordination. DMP encodes a trajectory in the form of a differential equation in phase space of the canonical system. Then these differential equations are integrated by a numerical integration method to reproduce the encoded trajectory. This work uses rhythmic DMP to produce locomotion in a virtual hexapod build in Matalb’s SimMechanics environment. Instead of using Receptive Field Weighted Regression (RFWR) as a function approximator, a neuro fuzzy inference system ELANFIS has been used. Use of ELANFIS in DMP’s framework has been given. Along with that a coupling architecture have been used to couple the canonical system, present in each leg, to achieve inter leg co-ordination. To validate control mechanism Hexapod is trained to perform three gait pattern namely, Tripod, Ripple, Wave. In each experiment Hexapod is trained with a gait pattern and left alone to reproduce the learned gait
