Gonzalo Tejera

Goal-oriented robot navigation learning using a multi-scale space representation

There has been extensive research in recent years on the multi-scale nature of hippocampal place cells and entorhinal grid cells encoding which led to many speculations on their role in spatial cognition. In this paper we focus on the multi-scale nature of place cells and how they contribute to faster learning during goal-oriented navigation when compared to a spatial cognition system composed of single scale place cells. The task consists of a circular arena with a fixed goal location, in which a robot is trained to find the shortest path to the goal after a number of learning trials. Synaptic connections are modified using a reinforcement learning paradigm adapted to the place cells multi-scale architecture. The model is evaluated in both simulation and physical robots. We find that larger scale and combined multi-scale representations favor goal-oriented navigation task learning.

Spatial cognition: robot target localization in open arenas based on rat studies

We describe our latest work in understanding spatial localization in open arenas based on rat studies and corresponding modeling with simulated and physical robots. The studies and experiments focus on goal-oriented navigation where both rats and robots exploit distal cues to localize and find a goal in an open environment. The task involves training of both rats and robots to find the shortest path to the goal from multiple starting points in the environment. The spatial cognition model is based on the rat’s brain neurophysiology of the hippocampus extending previous work by analyzing granularity of localization in relation to a varying number and position of landmarks. The robot integrates internal and external information to create a topological map of the environment and to generate shortest routes to the goal through path integration. One of the critical challenges for the robot is to analyze the similarity of positions and distinguish among different locations using visual cues and previous paths followed to reach the current position. We describe the robotics architecture used to develop, simulate and experiment with physical robots.

Learning Spatial Localization: From Rat Studies to Computational Models of the Hippocampus

In his landmark article, Richard Morris (1981) introduced a set of rat experiments intended “to demonstrate that rats can rapidly learn to locate an object that they can never see, hear, or smell provided it remains in a fixed spatial location relative to distal room cues” (p. 239). These experimental studies have greatly impacted our understanding of rat spatial cognition. In this article, we address a spatial cognition model primarily based on hippocampus place cell computation where we extend the prior Barrera–Weitzenfeld model (2008) intended to allow navigation in mazes containing corridors. The current work extends beyond the limitations of corridors to enable navigation in open arenas where a rat may move in any direction at any time. The extended work reproduces Morriss rat experiments through virtual rats that search for a hidden platform using visual cues in a circular open maze analogous to the Morris water maze experiments. We show results with virtual rats comparing them to Morriss original studies with rats.

An embedded particle filter SLAM implementation using an affordable platform

The recent growth in robotics applications has put to evidence the need for autonomous robots. In order for a robot to be truly autonomous, it must be able to solve the navigation problem. This paper highlights the main features of a fully embedded particle filter SLAM system and introduces some novel ways of calculating a measurement likelihood. A genetic algorithm calibration approach is used to prevent parameter over-fitting and obtain more generalizable results. Finally, it is depicted how the developed SLAM system was used to autonomously perform a field covering task showing robustness and better performance than a reference system. Several lines of possible improvements to the present system are presented.

Solving uncertainty during robot navigation by integrating grid cell and place cell firing based on rat spatial cognition studies

The efficient resolution of spatial localization is a key challenge in autonomous mobile robots. We describe in this paper our latest work in understanding spatial localization based on rat behavioral and neural studies. We develop a grid cell neural model based on studies in the Medial Entorhinal Cortex that integrates to a place cell neural model in the Hippocampus to generate “neural odometry” and spatial localization in the rat. We evaluate the model through simulated and physical robot experiments using a Khepera III autonomous robot in a laboratory environment.

Allothetic and idiothetic sensor fusion in rat-inspired robot localization

We describe a spatial cognition model based on the rats brain neurophysiology as a basis for new robotic navigation architectures. The model integrates allothetic (external visual landmarks) and idiothetic (internal kinesthetic information) cues to train either rat or robot to learn a path enabling it to reach a goal from multiple starting positions. It stands in contrast to most robotic architectures based on SLAM, where a map of the environment is built to provide probabilistic localization information computed from robot odometry and landmark perception. Allothetic cues suffer in general from perceptual ambiguity when trying to distinguish between places with equivalent visual patterns, while idiothetic cues suffer from imprecise motions and limited memory recalls. We experiment with both types of cues in different maze configurations by training rats and robots to find the goal starting from a fixed location, and then testing them to reach the same target from new starting locations. We show that the robot, after having pre-explored a maze, can find a goal with improved efficiency, and is able to (1) learn the correct route to reach the goal, (2) recognize places already visited, and (3) exploit allothetic and idiothetic cues to improve on its performance. We finally contrast our biologically-inspired approach to more traditional robotic approaches and discuss current work in progress.

A trial experience on management of MPLS-Based multiservice networks

This article presents a component-based, distributed management system for Multiprotocol Label Switched (MPLS) multiservice networks. Delivery of “triple play” multimedia services to the broadband residential user is a demanding challenge. The complexity is increased by the requirement of preserving Quality of Service (QoS) assurance for legacy connectivity services to the enterprise segment over the same infrastructure. New technologies are being introduced in the access, aggregation and core networks. Management applications must be aware of these advances and shall evolve accordingly. The proposed management architecture benefits from the capabilities of the MPLS Control Plane, in conjunction with a traditional management approach to provision QoS-aware services. This hybrid solution pursues short connectivity setup times by means of Control Plane signalling, with Traffic Engineering capabilities provided by the management framework. The system is being prototyped on a trial metropolitan testbed. Simulation results show that an advantageous trade-off between speed and resource optimisation is feasible.

A spatial cognition model integrating grid cells and place cells

Grid cells and place cells have shown to play an important role in spatial cognition in rats. While place cells provide global localization from external environment information, grid cells provide a “neural odometry” for path integration from internal vestibular information. In this paper we describe a spatial cognition model integrating grid cells and place cells from behavioral and neurophysiological brain studies in rats. Grid cell firing is generated from a linear oscillatory interference model that includes a reset mechanism to overcome errors in “neural odometry” readings. The model is evaluated in simulation. Future work is discussed including extensions to the model and evaluation under physical robots.

Computer Vision based System for Apple Detection in Crops.

In recent times there has been an increasing need to improve apple production competitiveness. The automatic estimation of the crop yield or the automatic collection may contribute to this improvement. This article proposes a simple and efficient approach to automatically detect the apples present on a given set of images. We tested the proposed algorithm on several images taken on many different apple crops under natural lighting conditions. The proposed method has two main steps. First we implement a classification step in which each pixel is classified as part of an apple (positive pixel) or as part of the background (negative pixel). Then, a second step explore the morphology of the set of positive pixels, to detect the most likely configuration of circular structures. We compare the performance of methods such as: Support Vector Machine, k-Nearest Neighbor and a basic Decision Tree on the classification step. A database with 266 high resolution images was created and made publicly available. This database was manually labeled and we provide for each image, a label (positive or negative) for each pixel, plus the location of the center of each apple.

Bio-Inspired Robotics: A Spatial Cognition Model integrating Place Cells, Grid Cells and Head Direction Cells

The paper presents a bio-inspired robotics model for spatial cognition derived from neurophysiological and experimental studies in rats. The model integrates Hippocampus place cells providing long-term spatial localization with Enthorinal Cortex grid cells providing short-term spatial localization in the form of “neural odometry”. Head direction cells provide for orientation in the rat brain. The spatial cognition model is evaluated in simulation and experimentation showing a reduced number of localization errors during robot navigation when contrasted to previous versions of our model.