|
| Learn More | Task Planning and Operations: Grasp Planning |
|
|
| Why study Grasp Planning and Mechanics? |
| Current Designs |
| Grasp Planning: Two schools of thought. |
| Our approach to Grasp Planning. |
| Videos
|
|
| Why study Grasp Planning and Mechanics? |
Grasp Mechanics becomes increasingly important when a robot must perform human-like grasping operations or employ tools designed to be used by humans as well as robots. Two separate issues are involved with the creation of an effective robotic hand. First, the duplication of human dexterity in a compact and reliable package is extremely difficult.
Second, while the job of grasping an object is usually accomplished with very little thought by most people, for a robot, grasping an object and maintaining that grasp requires a significant amount of knowledge about the local grasp environment and the the current interface between the hand and the object.
This information is very difficult to collect and process at an acceptable control rate. Further complicating this problem is the problem of dealing with the redundant appendages that grasping inherently entails, and the fact that engaging a grasp results in a serial to parallel transformation.
The human hand actually contains a total of 22-DOF contained in the thumb, four fingers, and the palm of the hand. This effectively results in a set of five 4-DOF serial fingers (robotic arms) mounted on a 2-DOF serial arm (the palm of the hand).
Several robotic hands have been developed over the years, becoming increasingly anthropomorphic or human-like in their design as the mechanical issues relating to the packaging of a reliable dexterous device in a suitable package have been addressed and the problems mitigated.
Many mechanical challenges still exist in terms of the stiffness and reliability of these hand designs, but development has reached the point where the hardware can support the development of appropriate control systems.
|
|
| Current Designs |
| Traditional Robotic Gripper |
| Stanford-JPL Hand |
| Utah-MIT Hand |
| Belgrade-USC Hand |
| Omni Hand |
| Hirzinger Hand |
| Robonaut Hand Conceptual Model |
|
| Grasp Planning: Two Schools of Thought |
In general, there are two schools of thought with respect to grasp planning, the empirical approach and the analytical approach.
The empirical approach is based on the abilities demonstrated by humans and animals. While modeling a mechanical system on the basis of animal systems may make a great deal of sense, this approach has proven to be exceptionally difficult due to our limited understanding of how animal systems actually accomplish these tasks.
By contrast, the analytical approach is physics based. Using suitable models of the hand, the environment and the laws that govern interactions in the universe, the analytical approach attempts to requirements for suitable grasps. However, except for special geometric cases, most of these solutions are computationally intractable.
These two approaches are complementary. Animal systems seem to employ analytical models to plan their grasps, and empirical approaches to confirm and maintain their grasp once engaged.
|
|
| Robotics Research Group Approach to Grasp Planning |
The Robotics Research Group takes advantage of this complementary approach to perform grasp planning and execution. User experience and instincts are used to place the hand in an appropriate position for grasping an object.
The hand then can use the available local model information provided by the user to choose a particular grasp configuration with the assistance of an appropriate set of grasp criteria.
Using the OSCAR Software system, the individual digits can be moved into position and the grasp can be initiated. Grasp initiation produces data from the hand sensors (Red Dots) that provides a positive contact indication.
This data can then be used to determine the current grasp quality. With this information available, the user can determine if the grasp is satisfactory, and if necessary repeat the grasp procedure.
|
|
|
