Vision enhanced neuro-cognitive structure for robotic spatial cognition

Abstract

This paper presents a brain inspired neural architecture with spatial cognition and navigation capability. The brain inspired system is mainly composed of two parts: a bio-inspired hierarchical vision architecture (HMAX) and a hippocampal-like circuitry. The HMAX encodes vision inputs as neural activities and maps to hippocampal-like circuitry which stores this information. Sensing a similar neural activity pattern this information can be recalled. The system is tested on a mobile robot which is placed in a spatial memory task. Among the regions in hippocampus, CA1 has place dependance response. With this property, the hippocampal-like circuitry stores the goal location according to the vision pattern, and recalls it when a similar vision pattern is seen again. The place dependent pattern of CA1 guides the motor neuronal area which then dictates the robot move to the goal location. The result of our current study indicates a possible way of connection between hippocampus and vision system, which will help robots perform a rodent-like behavior in the end.

Introduction

Animals such as rat and primates have a strong capability to navigate in a complex environment. Comparing to current SLAM algorithms [1], biological system continuously builds, maintains, and uses its spatial representations throughout its lifetime. In the meanwhile, without accurate geometric information, the biological system can effectively maintain a spatial representation suitable for path planning. Neurophysiological studies suggest that hippocampus, a major component of brain, plays an important role in memory and spatial navigation [2], [3]. The studies of neural activities in hippocampus in rat navigation experiments led to a theory that the hippocampus might act as a cognitive map: a neural representation of layout of the environment [4]. These findings arouse a great interest to incorporate the rat's navigation model in mobile robots navigation. Barrera et al. [5] proposed a neural structure to mimic “place field” property of hippocampus and guide a robot to search for the goal. The proposed system shows a similar learning pattern as rat experiments in the goal hunting [6]. Arleo and Gerstner used a joined CA1 and CA3 model to analyze the place cell property [7]. Reward-based learning is applied to map place cell activity into action cell activity to drive the motion. Wyeth and Milford [1] proposed a more biological based spatial cognition model which includes two parts: pose cells which function as grid cells in entorhinal cortex (EC) and experience map which functions as place cells in CA1 and CA3. In these research works, the focuses are the place cell property and how it contributes to the spatial navigation. The neural models are based on some sub-regions of hippocampus, without considering the hippocampus as a whole system.

Hippocampus has been studied for many years based on an approximate mammalian neuroanatomy. At a macroscopic level, highly processed neocortical information from all sensory inputs converges onto the medial temporal lobe where the hippocampus resides [8]. These processed signals enter the hippocampus via EC. Within the hippocampus [9], [10], [11], there are connections from the EC to all fields of the hippocampal formation, including dentate gyrus (DG), CA3 and CA1 through perforant pathway, from DG to CA3 through mossy fibers, from CA3 to CA1 through schaffer collateral, and then from CA1 back to EC. There are also strong recurrent connections within the CA3 region. Based on these connection properties of sub-regions, Edelman and his team [12], [13] developed a general brain-like model, namely brain-based device (BBD) to understand the hypotheses about how the mechanisms of the vertebrate nervous system give rise to cognition and behavior. Many interesting properties such as “place cells” and “episodic memory” in hippocampus have been realized with this model.

Inspired from the Edelman's BBD model, we target to develop a hippocampal-like cognitive system for the robot spatial navigation. In the BBD model, the vision inputs are only filtered by color and edge. As human, we can process very complicated vision information with various types of shapes. To improve the neural system's performance, we combine the HMAX model which is a bio-inspired vision process model and hippocampus circuity together. HMAX model has a unique feature when processes complicated shape information and it has scale invariance property. These properties help the neural system apply to a more complicated environment. The system also can help to understand how vision system and hippocampus work together in the spatial navigation.

The proposed system is mainly composed of two parts: hierarchical vision system and hippocampal-like circuitry. To explore the relationship between vision system and hippocampus, we implement the model in a mobile robot which is placed in a spatial memory task. The experiment environment is a plus maze where the robot is commanded to search for a hidden goal according to the vision. This maze environment has been used in rodent studies of spatial memory [14]. In the test, vision information is input to the HMAX model and its corresponding neural activities are mapped to the hippocampus. The region of CA1 shows a place-dependent response according to the vision inputs. This place-dependent pattern of CA1 guides the motor neuronal area which dictates the robot move to the goal location.