Abstract

Abstract

Abstract

Abstract
The continuous real-time motor interaction with our environment requires the capacity to measure and produce time intervals in a highly flexible manner. Recent neurophysiological evidence suggests that the neural computational principles supporting this capacity may be understood from a dynamical systems perspective: Inputs and initial conditions determine how a recurrent neural network evolves from a "resting state" to a state triggering the action. Here we test this hypothesis in a time measurement and time reproduction experiment using a model of a robust neural integrator based on the theoretical framework of dynamic neural fields. During measurement, the temporal accumulation of input leads to the evolution of a self-stabilized bump whose amplitude reflects elapsed time. During production, the stored information is used to reproduce on a trial-by-trial basis the time interval either by adjusting input strength or initial condition of the integrator. We discuss the impact of the results on our goal to endow autonomous robots with a human-like temporal cognition capacity for natural human-robot interactions.

Abstract
Neural field models, formalized by integro-differential equations, describe the large-scale spatio-temporal dynamics of neuronal populations [1]. They have been used in the past as a framework for modeling a wide range of brain functions, including multi-item working memory [2]. Neural field equations support spatially localized regions of high activity (or bumps) that are initially triggered by brief sensory inputs and subsequently become self-sustained by recurrent interactions within the neural population. We apply a special class of oscillatory coupling functions and analyze how the shape and spatial extension of multi-bump solutions change as the spatial ranges of excitation and inhibition within the field are varied [3]. More precisely, we use numerical continuation to find and follow solutions of neural field equations as the parameter controlling the distance between consecutive zeros of the coupling function is varied [4]. Important for a working memory application (e.g. [5]), we investigate how changes in this parameter affect the shape of bump solutions and therefore the maximum number of bumps that may exist in a given finite interval.

Abstract
Programming by demonstration allows non-experts in robot programming to train the robots in an intuitive manner. However, this learning paradigm requires multiple demonstrations of the same task, which can be time-consuming and annoying for the human tutor. To overcome this limitation, we propose a fast learning system - based on neural dynamics - that permits collaborative robots to memorize sequential information from single task demonstrations by a human-tutor. Important, the learning system allows not only to memorize long sequences of sub-goals in a task but also the time interval between them. We implement this learning system in Sawyer (a collaborative robot from Rethink Robotics) and test it in a construction task, where the robot observes several human-tutors with different preferences on the sequential order to perform the task and different behavioral time scales. After learning, memory recall (of what and when to do a sub-task) allows the robot to instruct inexperienced human workers, in a particular human-centered task scenario.