| Resumo: | The timing of movements and of action sequences is difficult when on-line coupling to sensory information is a requirement. That requirement arises in most behavior-based robot architectures, in which relatively low-level and often noisy sensor input is used to initiate and steer action. We show how an attractor dynamics approach to the generation of behavior in such architectures can be extended to the timing of motor acts. We propose a two-layer architecture, in which a competitive "neural" dynamics controls the qualitative dynamics of a second, "timing" layer. At that second layer, periodic attractors generate timed movement. By activating such limit cycles over limited time intervals, discrete movements and movement sequences can be obtained. We demonstrate the approach by simulating two tasks that involve control of timing: the interception of moving objects by a simple two-degree-of-freedom robot arm and the temporal coordination of the end-effector motions of two six-degree-of-freedom robot arms.
|
|---|