For some application areas, a robot needs to guarantee certain set of crucial tasks without having to do many kinds of tasks. Such areas include nuclear power plant applications, welfare robotics, and the rest of service robotics areas. For these applications, a robot needs to be designed so that it can guarantee to achieve only a certain set of pre-specified tasks. This approach, called task-oriented design (TOD), is expected to attract increasing attentions in step with the growth of service robotics.
The design procedure for task-oriented robots can be divided into two parts as follows: kinematic and dynamic design. As can be seen from this division, kinematic design is very important in that it is the first stage of the whole design procedure for a robot manipulator. A small mistake in this stage could easily make efforts in subsequent stages almost useless, however successful they may be as such. In this paper, therefore, I focus on optimum kinematic design of a robot manipulator in the context of TOD.
There have been various researches on optimum design of robot kinematics. However, there have been a few researches on the task-oriented design procedure for robot kinematics itself. Therefore, I divided kinematic design procedure into a sequence of stages and described each stage in detail , providing a systematic and effective design procedure for robot optimum kinematics.
The core design stage for task-oriented design of robot kinematics is the optimum design stage, which consists of two steps as follows: (i) problem formulation which sets design variables, cost function and constraints, and (ii) optimum solution search for the formulated problem using one of the various local or global optimization techniques. Of these two, more important step is the former, i.e. how the design problem is mathematically formulated.
To reduce the complexities and increase the efficiency in the previous problem formulation methods, Yang(2000) introduced the concept of Grid Method, which is widely used method in the finite difference method or numerical analyses of heat transfer and fluid flow, into the design area of robot kinematics. Despite its conceptional merits, using wrong and non-systematic formulation which was not based on the optimum design theory gave Yang's unit workspace griding method the following problems: (i) lots of computational burden (ii) being applicable to kinematic design of planar manipulator only (iii) local optimization.
To solve these problems, I newly proposed Grid Method based on the right concept of Grid Method and the optimum design theory. Grid Method selects joint position and twist angle as design variables. Since the number of design variables remains constant, i.e. '4', unlike previous works, the number of DOF and task points does not affect the complexities of the problem formulation. This implies that Grid Method becomes more efficient as the number of DOF and task points both increase, considering that the search space in general increases exponentially as the number of design variables increases. Due to this, the cost function and constraints can be represented very simply. In addition, Grid Method can easily calculate DH parameters from the geometric structure between links of unit grid without solving the inverse kinematics, because it selects joint position and twist angle as design variables. Furthermore, it does not require forward kinematics, because the positions of the end-effector are initially fixed to those of the task points to satisfy the boundary conditions.
The stage following the problem formulation is to find an optimum solution using one of local or global optimization techniques. To solve the problem that the unit workspace griding method finds a local optimum only, I provided Very Fast Simulated Annealing, a global optimization technique, as well as Generalized Reduced Gradient Method, a local optimization technique. Finally I proposed the whole optimum design algorithms by joining the problem formulation using Grid Method and the optimization techniques above.
In order to make the solution converge into an optimum, we should tune the weights included in each term of the cost function. To this end, I proposed an adaptive weights tuning algorithm to increase the efficiency of finding an optimum as well as a manual weights tuning algorithm using fixed weights.
So far I proposed systematic and effective task-oriented design procedure for robot optimum kinematics, showing its effectiveness through applications to various robot manipulators. When we realize the optimum design procedure for robot kinematics as proposed in this paper by a computerized program, only initial posture of robot manipulator for every task point is required to automatically produce robot optimum kinematics. By virtue of this, even a novice designer can easily design robot kinematics and the development period of robot is expected to be shortened.