This paper gives a comprehensive study on “computational robotics” motivated by the recent rapid development of supercomputer technologies. By applying advanced computational scientific approach, it is expected to realize deeper understanding of human as well as to solve more complicated computational problems for the next generation of robotics. Specifically, three main problems are discussed. The first one is about high dimensional super-redundancy considering the full body human motor functions. With the rapid computation ability, it will be possible to realize simultaneous human movement measurement with dynamic human motion analysis and simulation within real time. The second topic is on the massive computation of a robot's complex full body physical interactions with environment and/or manipulated objects such as a cared person. Here, how to simulate, plan and control the interactions with many kinds of frictions are discussed. The discrete feature of physical contacts together with continuous behaviors of motion/force control makes the problem as a hybrid dynamic system. Systematic approach to solve such problem requires massive computation. The final one simply studies on a robot's cognitive motion in the real world so that to adapt to the unknown and/or uncertain environment. Such problem requires to process huge cognitive information inputs within real time so as to determine the robot's environmental adaptive actions.