|
  |
|
|
Krasnow Institute > Monday Seminars > Abstracts Machine psychology: Jeff Krichmar Models of neural function in computational neuroscience are typically simulated on a computer that is detached from the world and receiving artificially generated inputs. There are a couple of limitations to this approach. First, since the inputs to the model or the environment are simulated, the world model is not as rich as a noisy real-world environment. In addition, modelers may bias their model by inadvertently leaving out details or simplifying their model of the environment. Second, the nervous system is attached to a body that interacts with its environment. The morphology and phenotype of the organism is crucial for the development and operation of the nervous system. The central role of sensorimotor activity provides an important rationale for "embodied" models of neural function, i.e. models that include movements of sensors and effectors. We have developed a paradigm in which a Neurally Organized Mobile Adaptive Device (NOMAD) explores its environment and through learning is able to develop adaptive behaviors. NOMAD's behavior is guided by a simulated nervous system based roughly on the anatomy and physiology of the mammalian brain. There are six major components in its nervous system: a visual system, an auditory system, a taste system, sets of motor neurons capable of triggering behavior, a tracking system driven by visual stimuli, and a value system. NOMAD consists of a mobile base, equipped with sensors for vision, hearing and taste, and effectors for movement of its body, of its head and of a one degree-of-freedom gripping manipulator. Using this platform, we have pointed out the importance of: 1) individual history in an environment for the development of perceptual categories, 2) a modulatory system to transfer the value sensed by one sensory system to another, and 3) the notion that the brain has different learning rates for punishment and reward and that this difference is crucial for the organism's survival. These results would not be possible without the implementation of a real-world device behaving in its environment and responding to a simulated nervous system. These experiments reinforce the importance of testing nervous system theories on embodied platforms.
The Krasnow Institute for Advanced Study |
|
