Currently, we refer to the basic actions in Jack as the primitive actions. These include actions such as move (which includes translate and rotate within it), grasp, reach, locomotion, visual search, task-appropriate attention, etc. Each of these primitive actions can be generated by various methods like motion scripting, live motion capture, various algorithms using the principles of dynamics and control, fast inverse-kinematics, random noise, walking algorithms, etc. Instead of generating motions for the whole agent (we consider only the human agent for this purpose) using only one of these sources at a time, we could use a different source for each part of the body. For this purpose, currently, at a higher level, the body of the human model is divided into the following parts - head, arms, torso, legs, and figure position. A different PaT-Net is used for each motion generator. The parameters to the motion generator would then be the part of the body that they would control. If more than one motion generator were to control the same part of the body, it could be resolved in two ways - blend (or any other binary operation) the resulting motions together or set up a priority for the motion generators and arbitrate appropriately between them.