Abstract
These works deal with the computation of optimal motions for the humanoid robots. Most of the motion planning methods come from the motion planning of the manipulator robots. They rely on optimization algorithms which need a motion parametrization and a time-discretization of the constraints that define the physical limits of the robot. We show that a time-grid discretization is hazardous for the safety and the integrity of the robot. That is why, we propose a new method for the guaranteed discretization that computes the extrema of the constraints over time-interval that covers the whole motion duration. This method of discretization is time consuming. Thus, we developped a hybrid method that ensures the constraint validity within the same range of time of the state-of-the-art methods. With this method, we created a database of motions to follow a moving target. Consequently, we can generate an optimal motion that fits to the environment. However, there is no method which is fast enough to compute a new motion adapted to a new environment. Thus, we present a re-planning method that produces a new motion from a previous one. To do it, we compute, offline, a feasable sub-set around the motion that respects the constraint validity. The re-planning process consists in finding, in this sub-set, a new motion that is adapted to the new environment. We tested this re-planning method with a kicking motion where the position of the ball changes and we are able to find and adapted motion within 1.5s of CPU-time.