Robotic manipulation using a behavioral framework

We endeavor to build a robotic manipulation system that is capable of functioning in an unstructured environment. In order to provide such functionality, the system must be capable of reacting in a compliant manner to any stimuli in the environment at the motor level, and use the same information to modify its existing plan of action. To build a general enough system that can handle a wide range of manipulatory tasks we build motion primitives both at the task level, which we term motor programs, and at the motor level which we term reflexes. Motor programs specify the behavior of the system at the motor level in terms of the desired stimuli response characteristic. Thus we introduce the concept of behavior to represent the sensorimotor relationship that can be defined at different levels of detail. More complex behaviors are manifested by coordinating the behavior of a number of motor systems. Borrowing extensively from advances in the area of neurobiology and neuropsychology, we mimic the human motor control system by creating a hierarchical control system in which we incorporate the motor programs at the highest level. At the lower levels we breakdown the specified behavior into in constituent reflex actions for each of the manipulators in the system. Between the task level and the motor level we create a planning level which adaptively sequences through the motor programs, and a sensorimotor level that reactively coordinates the behavior of the manipulators in the system to produce intelligent behavior. The behavioral robotic framework is used both on a graphics simulator as well as an experimental testbed. The testbed consists of two robot arms, one mounted with an articulated mechanical hand instrumented with tactile sensors, and the other mounted with a mobile range scanner. We test, verify and extend the hypothesis that complex actions can be accomplished using simpler but coordinated reflexive motions