In this thesis, are investigated the crab gait of a quadruped and its application on adaptive-gait-control problem. Some important characteristics that simplify the analysis of crab gaits are presented, and several formulas for optimizing the longitudinal gait stability margin of a quadruped crab gait are derived by incorporating the time weighting factor. A unique gait is also suggested which optimizes gait stability margin according to the range of crab angle. Further, is considered effects of variations of footholds in stability and maximum permissible stroke in terms of support boundary angle.
Many previous works in gait analysis employ a moving body coordinate frame as a reference frame for specifying the position of foothold. But the foothold of a leg has only two discrete values in one locomotion cycle. Which are present foothold and next foothold. This simple observation allows us to use a "pseudo" world coordinate frame as a reference frame. By employing this frame, the analysis and implementation of the gait have been greatly simplified.
The results regarding the basic characteristics and the stability properties for the quadruped contain the previous works on the forward walking gait as a special case of the crab gait.
On the basis of the above results, is considered the problem of adaptive-gait-control for a quadruped with particular attention to the issue of eliminating dead-lock positions. The adaptive-gait-control problem, which is considered quiet complex in modeling and in finding a solution, is formulated in the form of a constrainted optimization problem. Then a solution approach is proposed based on a neural optimization network. The desired footholds are taken as the output variables of the neural circuit and differential equations are derived where solutions tend to minimize the energy of the network. An efficient method for making the quadruped statically stable is addressed. Several conditions for satisfying the kinematic limits of reachable volumes are discussed. The concept of mobility is incorporated into the energy function of the network for an efficient adaptation to the next motion trace. It is shown that the output of the network converges to the wave-crab gait in the case of no obstacle, and consequently the algorithm is applicable to both even and rough terrains without gait transitions. Further, it is demonstrated that the neural optimization network can generate good solutions for the adaptive-gait-control problem.
