Key points are not available for this paper at this time.
The paper reports on a project to make a quadruped robot walk with medium forward speed on irregular terrain in an outdoor environment using a neural system model. The necessary conditions for stable dynamic walking on irregular terrain in general are proposed, and the neural system is designed by comparing biological concepts with those necessary conditions described in physical terms. A PD-controller is used at joints to construct a virtual spring—damper system as the visco-elasticity model of a muscle. The neural system model consists of a CPG (central pattern generator), responses and reflexes. A response directly and quickly modulates the CPG phase, and a reflex directly generates joint torque. The state of the virtual spring—damper system is switched, based on the CPG phase. In order to make a self-contained quadruped (called Tekken2) walk on natural ground, several new reflexes and responses are developed in addition to those developed in previous studies. A flexor reflex prevents a leg from stumbling on small bumps and pebbles. A sideways stepping reflex stabilizes rolling motion on a sideways inclined slope. A corrective stepping reflex/response prevents the robot from falling down in the case of loss of ground contact. A crossed flexor reflex helps a swinging leg keep enough clearance between the toe and the ground. The effectiveness of the proposed neural system model control and especially the newly developed reflexes and responses are validated by indoor and outdoor experiments using Tekken2. A CPG receives sensory feedback as a result of motions induced by reflexes, and changes the period of its own active phase. Since a CPG has the ability of mutual entrainment with pitching motion of legs and rolling motion of the body in addition, the consistency between motion of a leg temporally modified by a reflex and motions of the other legs is maintained autonomously. It is shown that CPGs can be the center of sensorimotor coordination, and that the neural system model simply defining the relationships between CPGs, sensory input, reflexes and mechanical system works very well even in complicated tasks such as adaptive dynamic walking on unstructured natural ground.
Kimura et al. (Tue,) studied this question.