The aim of this project is to explore high speed decision making by a humanoid robot for motion generation. Humans, such as firemen or sportsmen, are able to take a pertinent decision in the blink of an eye although such decision usually involves a rather time consuming deliberative process. The so called snap-judgments are usually obtained after training, practice and an extended experience of similar situations. Robots interacting with humans performing collaborative work for instance, face the same kind of challenges. Indeed let us imagine a humanoid robot manipulating a table with a human (such as demonstrated in 2003 by the Humanoid Research Group of AIST ). If the latter one loses grip of the table, the robot will have to quickly move appropriately to avoid putting in danger its human collaborator and itself. Finding in a timely manner a safe sequence of motion to avoid such situation is crucial. There are other field of applications, less dramatic, where such capabilities would be useful such as entertainment where a small-size robot could imitate a child, or interact with him during a game. In such context, obviously it would not be possible to assume that the user is an expert able to program the robot appropriately.