Brain Area V6A: A Cognitive Model for an Embodied Artificial Intelligence

We found that single neurons in the parietal area V6A of the macaque brain deal with all the components of reaching and grasping actions: locating in space the object target of action, directing the eyes toward it, sensing where the arm is in space, directing the arm toward the spatial location where the object is in order to reach and grasp it, adapting the grip to the object shape and size. The knowledge of how the brain codes simple visuomotor acts can be useful to build artificially-intelligent systems that have to interact with objects, localize them, direct their arm toward them, and grasp them with their gripper. Single cell recordings can also be useful in understanding how to perform more complex visuomotor tasks, like interacting with human beings, exchanging objects with them, and acting in an ever changing environment. An humanoid robot gazes at a dish that it should insert into a dishwasher. The robot reaches the dish with its arm and grasps it with its gripper, lifts the dish, and puts it into the appropriate place on the dishwasher plate. The robot has produced a sequence of actions that every human being performs naturally and dexterously hundreds of times each day: locating a visual object in space, directing the eyes toward it, directing our arm toward the spatial location where the object is in order to reach and grasp it, adapting the grip to the object shape and size. In order to perform this task successfully, our brain, and also the cognitive architecture of a robot, should know where the eyes are directed, where is its hand in space, and where is located the goal of action is in peripersonal space. The prehension task is achieved by primates through a series of neural elaborations that are performed in the parietal and frontal lobes. Recording of spike trains from single neural cells and analysis of their modulations according to the different phases of the prehension task are the most used techniques through which we acquire knowledge of how the prehension task is achieved. This is the job of neurophysiologists (like us) who select a brain area (supposed to be involved in certain functions), and record the bioelectrical signals from single cells of that area with fine wire microelectrodes. The frequency of discharge of action potentials (spikes) changes according to the signal that is processed by the neuron itself. The knowledge on how the brain codes the different phases of prehension task can be useful to build up artificially-intelligent systems, in particular to build embodied aspects of cognition. So we propose here a summary of our studies in order to solve some problems that scientists who are involved in artificial intelligence (AI) could encounter. Neurophysiology of Prehension We cannot record from the human brain (or can only occasionally, and for a very short time, during a neurosurgery), because of ethical reasons. If we want to know how the human brain works, we have to study the brain of an animal that is able to perform the same task we want to investigate in human. For studying the brain control of prehension, the most used animal is the macaque because its visuomotor functions are almost identical to the human being. We have been studying for several years a region of the macaque brain known to be involved in visuomotor functions. In particular, we are currently studying the functional properties of neurons of a parietal area called V6A (V stands for visual, as