Research

Intelligent Soft Arm Control

Major System Modules
ISAC has a distributed object-oriented architecture. It uses a blackboard scheme to communicate among system modules.

• Task planner. Task planning in ISAC is achieved in a totally distributed fashion. It is performed by several task decomposition agents that interact via blackboard to decompose high level user commands into primitive actions.
• Parallel control. Soft Arm is controlled by a transputer-based parallel controller. It uses a network of transputers that can be reconfigured in case of a fault in the controller.
• Reflex control. ISAC has a reflex control capability to insure user safety in case the user moves suddenly towards the robot arm. In this case, the real-time face tracking system automatically moves the robot arm to a safe position.
• Voice system. The voice system is used to replace the keystroke as a main user interface. Currently we are using the Phonetic Engine 500 from Speech Systems, Inc. on a Windows PC. It is running in parallel with the planning process, allowing the user to intervene the arm execution if necessary.
• Real-time face tracking. Real-time face tracking is done for collision prediction and avoidance. We have developed a face tracking algorithm which recognize a human face in the 3-D space and tracks the face in real time. It is running concurrently with task planning and reflex control.
• Object recognition. We have developed a fast size and rotation invariant 2-D object recognition system using histogram and the log-polar transform. The module can recognize cups and plates and various utensils such as forks, spoons and knives.

Current Research
Currently we are extending ISAC capabilities by adding another 6 DOF Soft Arm to the system. This will enable us to perform more complex tasks for the user which requires cooperating arms. This new arm will be equipped with a flexible robotic hand for picking up complex or fragile items (e.g. potato chips, etc.). Software modules for ISAC under development include:

• Action/Macro Builder. This module acts as a voice activated "teach pendant" for the system designer or user. It provides the user with the ability to teach ISAC new actions and later retrieve them. This module enhances the extensibility of the ISAC's tasks.
• Learning Module. The idea is to add a self learning capability to ISAC in order to enable the system in "reasoning" about primitive actions it must executes under a high level task command. The system then will be capable of assimilating new tasks automatically.
• Neural Net and Fuzzy Control. We are developing a control system which can learn the best control strategy using neural network and fuzzy logic. The neural network will be used to generate the knowledge base which will be used by the fuzzy controller.

All IRL students are involved to some varying degree in the ISAC project.