A Phase/Event/Transition Framework for Dextrous Manipulation

In order to simplify the complexity of using multiple fingers and the multiple control laws that are often necessary to perform a task, the DML has developed a framework for event-driven dextrous manipulation.

A dextrous manipulation task can be viewed as a sequence of phases punctuated by events. For example, as the fingers of a hand close upon an object they are driven using position control, but when manipulating an object they are driven to maintain control of internal forces in the object. In this example, the sensation of contact is an event that signals the transition from one control phase to the next.

The following diagram shows an example of a simple Aquire/Grasp/Lift/Replace task decomposed into phases and events. Top level Manager Phases control the execution of chains of Action phases. Some of the action phases are "coordinated motion phases" while others, such as the Acquire phase, involve parallel sequences of independent finger motion. Each phase has an associated control law and force/motion trajectory, and one or more expected terminating events. At the symbolic level, this is a Discrete Event System, and could be modelled using Petri nets. At a finer level, the details of controls and transitions become important as shown in the following pages.

The figure below shows a transition from an Approach phase to a Contact phase in more detail. Here, the controller must switch from motion to force control. The Alert confidence for the contact event shown below allows the controller to ramp down velocity gains in order to prepare for the transition to force control. A smooth transition is important for dextrous manipulation in order to avoid the excitation of dynamic tactile sensors.

Details on the control framework and controller design can be found in:

• [Hyde, Tremblay and Cutkosky], "An Object-Oriented Framework for Event-Driven Dextrous Manipulation", 4th Intl. Symposium on Experimental Robotics

• Jim Hyde's PhD thesis