BookPDF Available

Robot Navigation from Nature: Simultaneous Localisation, Mapping, and Path Planning Based on Hippocampal Models

Authors:

Abstract and Figures

This thesis describes the use of a hippocampal model as a basis for a complete robot mapping and navigation system. Computational models of animal navigation systems have traditionally had poor practical performance when implemented on robots. The aim of the work was to determine the extent to which hippocampal models can be used to provide a robot with functional mapping and navigation capabilities. The thesis starts with a review of the mapping and navigation problem and the robotic and biological approaches to providing a solution. The problem is a broad one and to completely solve it a robot must possess several key abilities. These include the ability to explore an unknown environment, perform Simultaneous Localisation And Mapping (SLAM), plan and execute routes to goals, and adapt to environment changes. The most successful conventional solutions are based on core probabilistic algorithms and use a variety of map representations ranging from metric to topological. These probabilistic methods have good SLAM performance, but few have been successfully integrated into a complete solution to the entire mapping and navigation problem. In contrast, biological systems are generally poorly understood and current models emulating them have poor practical performance. However, many animals solve the entire mapping and navigation problem without the precise sensors and high resolution maps that are typical of robotic methods. Models of the mapping and navigation process in relatively well understood animals such as rodents offer much potential for practical improvement. The thesis describes a series of studies which implemented computational models inspired by the mapping and navigation processes in the rodent hippocampus. The initial study was based on conventional theories of separate orientation and location representations in the rodent hippocampus. The model was tested in a range of experiments that revealed fundamental limitations in its practical usefulness as a SLAM system. A review of the literature revealed no strong evidence suggesting that rodents overcome these limitations and no mechanisms by which they might do so. Consequently the model was then modified to combine its separate representations of robot orientation and location into a single representation of robot pose. This new model, known as RatSLAM, was able to successfully perform SLAM on two different robot platforms in a range of indoor and outdoor environments. A goal memory system was added to RatSLAM to give the robot the ability to plan and execute routes to goals. However, in large complex environments the RatSLAM model’s pose representation exhibits a number of phenomena that cause the goal memory system to fail. These phenomena were removed or modified by the introduction of a new experience mapping algorithm. This algorithm uses the RatSLAM maps to create spatio-temporal-behavioural representations known as experience maps, which preserve the topological structure of the original map. When used in a number of indoor and outdoor experiments the algorithm generated representations that were globally topological and locally spatial. Subsequent work implemented methods for exploration, goal recall, and adaptation using the representations built by the RatSLAM model and experience mapping algorithm. Each of these processes was tested in a range of autonomous robot experiments. In the final experiments the robot, acting in a completely autonomous manner, explored an unknown environment while performing SLAM, then navigated to goal locations while adapting to simple lasting environment changes. The studies described in this thesis provide a detailed analysis of the mapping and navigation capabilities of current computational models of biological systems within the context of producing functional robot systems. Together the RatSLAM model and experience mapping algorithm bring a biologically inspired method into the realm of conventional robot mapping and navigation techniques for the first time. The combined system was developed heuristically and displayed successful, although sometimes suboptimal, performance in a range of real robot tests lasting up to two hours. The work stands as one of the few examples of a complete integrated attempt to solve all the major mapping and navigation challenges faced by mobile robots.
Content may be subject to copyright.
A preview of the PDF is not available

Chapters (11)

Mobile robots and animals alike require the ability to move around in their environments. For a large number of animals such as insects, this ability is provided by reactive movement schemes. With an appropriate set of reactive behaviours an insect can function quite well in complex environments. However, there is a limit to what can be achieved through pure reactivity. A whole class of navigation problems become solvable if an animal or robot is able to form some sort of map of its environment.
Sensor and environment uncertainty has caused most robotic mapping and navigation methods to converge to probabilistic techniques. The key strength of probabilistic techniques is their ability to deal with uncertainty and ambiguity in a robot’s sensors and environment. Any technique that has some means of handling the uncertainties faced by a mobile robot has an immense advantage over techniques that do not. Probabilistic techniques allow a robot to appropriately use sensory measurements based on their modelled uncertainties. Ambiguous features or landmarks in the environment become useful (albeit less useful than unique features) rather than becoming failure points for the mapping algorithm.
Many animals possess formidable navigation abilities. Conducting research into the exact capabilities and limitations of the mapping and navigation systems used in nature provides two valuable outcomes; knowledge of what animals and insects can achieve using their own sensory and computing equipment, and theories and models of the mechanisms they use in the process of doing so.
This chapter covers the state of the art models of the rodent hippocampus and their ability to solve each component of the mapping and navigation problem. Rodent hippocampal models can be separated into two groups. The first contains models that have been developed using theories of the neuronal mechanisms by which rats navigate and create a map. By testing these models in simulation or on a robot these theories can be validated or disproved. These models can also be used to make predictions that can be tested by further rodent experiments. Most of the models described in this chapter fall into this group.
The last four chapters have discussed the mapping and navigation problem and the range of system solutions that exist both in the animal world and on robot platforms. There are many forces driving the diversity that is apparent when examining these systems. In nature, creatures and their navigation systems have evolved to suit their environments and their own physical and sensory characteristics. In the natural world, the range of body types, sensory equipment, environments, and lifestyles has produced a myriad of solutions to the problems facing a creature that needs to move around effectively in its environment. Likewise in the research labs of robotics researchers and the domestic home or industrial workplace, mapping and navigation systems have developed in ways to suit the environments, the sensors available and the purpose of the robot.
Rodents are the most thoroughly understood animal with respect to both their mapping and navigation capabilities and the neuronal mechanisms by which they are achieved. However, of the few studies that have actually implemented a hippocampal model on a robot (Arleo 2000; Krichmar, Nitz et al. 2005), none have investigated the full extent to which the model can be used as a practical mapping and navigation system. This chapter presents the implementation and experimental testing of a classical model of the mapping and navigation processes in the rodent hippocampus.
Separate representations of robot orientation and spatial location are inherently unsuitable for mapping and navigation in large, ambiguous environments, as demonstrated by the work presented in Chapter 7. This chapter describes the implementation of an extended hippocampal model known as RatSLAM, which combines the concept of head direction and place cells to form a new type of cell known as a pose cell.
SLAM methods vary significantly, which can make direct comparison challenging. However there is one performance indicator which can be universally applied to all these methods – can they be used to perform useful tasks such as goal orientated navigation? Navigating to goals is one of the fundamental purposes of any mapping and navigation system, whether it is one used by a rat foraging for food, or by a delivery robot in an office environment. The type and nature of the goal may vary between contexts, but the core problem usually remains the same; how can the animal or robot efficiently get from place A to place B. This chapter describes the implementation of a method for goal navigation using the RatSLAM system and a cell structure known as goal memory.
In environments that are larger than the nominal area encoded by the pose cells, the RatSLAM model reuses the pose cell matrix multiple times in creating a map of the environment. This leads to several phenomena in the pose cell matrix: discontinuities where adjacent physical areas are represented by distally separate cells; multiple representations where several clusters of cells represent the same physical area; and collisions where several physical areas are represented by the same set of cells. Although the representations are topologically correct, consistent and stable, these phenomena remove the spatial relevance of the pose structure layout and hence any possibility of using it directly for tasks such as goal navigation. One approach to overcoming this problem is to create an algorithm that can maintain the inter-cell spatial information while retaining the topological correctness of the pose cell maps. The experience mapping algorithm was developed to pursue this approach.
The experience mapping algorithm uses a process of map correction to minimise the discrepancies between the locations of experiences in the map and the spatial information stored in the inter-experience links. Links between experiences also contain temporal and behavioural information. When combined with the spatial information, this temporal and behavioural information can be used to explore an environment, navigate to goals, and adapt to changes in the environment.
This book has described the implementation of a vision-based simultaneous localisation and mapping system using extended models of the rodent hippocampus, and an experience mapping algorithm that integrates with this system to provide exploration, goal recall and adaptation capabilities. The purpose of this chapter is to summarise the work presented, discuss the research contributions, and suggest potential future work in this field.
... Milford et al. made significant pro gress in emulating the spatial navigation ability of the hippocampal system by building a semimetric topological map in a very large area (Milford andWyeth 2008, 2010). In their work, the core model, RatSLAM, has been demonstrated to build maps si mul ta neously in large and complex environments. ...
... • This book is the key publication presenting the hippocampal-based approaches to robot navigation and the RatSLAM approach: Milford, Michael. 2008. Robot Navigation from Nature: Simultaneous Localisation, Mapping, and Path Planning Based on Hippocampal Models. Vol. 41. Berlin: Springer Science and Business Media. ...
Chapter
Full-text available
The current state of the art in cognitive robotics, covering the challenges of building AI-powered intelligent robots inspired by natural cognitive systems. A novel approach to building AI-powered intelligent robots takes inspiration from the way natural cognitive systems—in humans, animals, and biological systems—develop intelligence by exploiting the full power of interactions between body and brain, the physical and social environment in which they live, and phylogenetic, developmental, and learning dynamics. This volume reports on the current state of the art in cognitive robotics, offering the first comprehensive coverage of building robots inspired by natural cognitive systems. Contributors first provide a systematic definition of cognitive robotics and a history of developments in the field. They describe in detail five main approaches: developmental, neuro, evolutionary, swarm, and soft robotics. They go on to consider methodologies and concepts, treating topics that include commonly used cognitive robotics platforms and robot simulators, biomimetic skin as an example of a hardware-based approach, machine-learning methods, and cognitive architecture. Finally, they cover the behavioral and cognitive capabilities of a variety of models, experiments, and applications, looking at issues that range from intrinsic motivation and perception to robot consciousness. Cognitive Robotics is aimed at an interdisciplinary audience, balancing technical details and examples for the computational reader with theoretical and experimental findings for the empirical scientist.
... When investigating the way rodents navigate from a bioinspired perspective, RatSLAM as introduced in [7], [10], [13]- [15], has been proven to be a capable visual SLAM method. RatSLAM is loosely based on the neural processes underlying navigation in the rodent (primarily rat) brain, more specifically the hippocampus. ...
... The pose cells [7] encode the robot's current best pose estimate. Pose cells are represented by a Continuous Attractor Network (CAN) [13,Ch. 4], the posecell network, to resemble the grid cells as introduced in [16]. ...
Preprint
RatSLAM is a rat hippocampus-inspired visual Simultaneous Localization and Mapping (SLAM) framework capable of generating semi-metric topological representations of indoor and outdoor environments. Whisker-RatSLAM is a 6D extension of the RatSLAM and primarily focuses on object recognition by generating point clouds of objects based on whisking information. This paper introduces a novel extension to both former works that is referred to as ViTa-SLAM that harnesses both vision and tactile information for performing SLAM. This not only allows the robot to perform natural interaction with the environment whilst navigating, as is normally seen in nature, but also provides a mechanism to fuse non-unique tactile and unique visual data. Compared to the former works, our approach can handle ambiguous scenes in which one sensor alone is not capable of identifying false-positive loop-closures.
... By contrast, later models aimed to provide metric scene representations, with metric information either the explicit input [207] or the implicit output [208, p.107], [209]. Halfway between them, RatSLAM [210][211][212][213] aimed for a fine grained topological map that could facilitate shortcuts, but was less than fully metric ('the map does not follow a strict Cartesian coherence' [210]). ...
Article
New approaches to 3D vision are enabling new advances in artificial intelligence and autonomous vehicles, a better understanding of how animals navigate the 3D world, and new insights into human perception in virtual and augmented reality. Whilst traditional approaches to 3D vision in computer vision (SLAM: simultaneous localization and mapping), animal navigation (cognitive maps), and human vision (optimal cue integration) start from the assumption that the aim of 3D vision is to provide an accurate 3D model of the world, the new approaches to 3D vision explored in this issue challenge this assumption. Instead, they investigate the possibility that computer vision, animal navigation, and human vision can rely on partial or distorted models or no model at all. This issue also highlights the implications for artificial intelligence, autonomous vehicles, human perception in virtual and augmented reality, and the treatment of visual disorders, all of which are explored by individual articles. This article is part of a discussion meeting issue ‘New approaches to 3D vision’.
... Local view cells are primarily used to store visual templates, which are indispensable for map construction and positioning systems [23]. There are two types of interactions between local view cells and pose cells simultaneously: map construction by associative learning from each other and localization update by injecting activities into pose cells [24]. ...
Article
Full-text available
Inspired by rodents’ free navigation through a specific space, RatSLAM mimics the function of the rat hippocampus to establish an environmental model within which the agent localizes itself. However, RatSLAM suffers from the deficiencies of erroneous loop-closure detection, low reliability on the experience map, and weak adaptability to environmental changes, such as lighting variation. To enhance environmental adaptability, this paper proposes an improved algorithm based on the HSI (hue, saturation, intensity) color space, which is superior in handling the characteristics of image brightness and saturation from the perspective of a biological visual model. The proposed algorithm first converts the raw image data from the RGB (red, green, blue) space into the HSI color space using a geometry derivation method. Then, a homomorphic filter is adopted to act on the I (intensity) channel and weaken the influence of the light intensity. Finally, guided filtering is used to process the S (saturation) channel and improve the significance of image details. The experimental results reveal that the improved RatSLAM model is superior to the original method in terms of the accuracy of visual template matching and robustness.
... These efforts have yielded impressive results. For example, RatSLAM [20]- [23] has been shown to be able navigate office spaces [24] and map large outdoor environments [25]. The algorithm has also been used successfully with multiple sensors (including cameras, sonar, and electromagnetic sensors) and to support sensor fusion [26]- [28]. ...
Article
Estimating the fastest trajectory is one of the main challenges in autonomous vehicle research. It is fundamental that the vehicle determines its path not only to minimize travel time, but to arrive at the destination safely by avoiding any obstacles that may be in collision route. In this paper, we consider estimating the trajectory and acceleration functions of the trip simultaneously with an optimization objective function. By approximating the trajectory and acceleration function with B-splines, we transform an infinite-dimensional problem into a finite-dimensional one. Obstacle avoidance and kinematic constraints are carried out with the addition of a penalization function that penalizes trajectories and acceleration functions that do not satisfy the vehicles’ constraints or that are in a collision route with other obstacles. Our approach is designed to model observations of the obstacles that contain measurement errors, which incorporates the realistic stochasticity of radars and sensors. We show that, as the number of observations increases, the estimated optimization function converges to the optimal one where the obstacles’ positions are known. Moreover, we show that the estimated optimization function has a minimizer and that its minimizers converge to the minimizers of the optimization function involving the true threat zones.
Thesis
Full-text available
Citation: Fangwen Yu. Brain Inspired SLAM System for 3D Environments [D]. Wuhan, China: China University of Geosciences, 2020: 1-218. Abstract: With the development and widespread applications of land, sea, air and space-based robots, the autonomy, robustness and intelligence of the robot system may face new grand challenges. How to make robots truly autonomous has become a popular research topic. As one of the key technologies to realize robot autonomy, three-dimensional (3D) localization and mapping (SLAM) still faces many challenges in practical applications under unknown, unpredictable, complex, 3D environments. Furthermore, there are strictly limitations on the robot, such as the on-board sensors, computing, power consumption, storage, weight, size, etc. In addition, sensor signals are susceptible to interference and the system cannot rely on external signals such as GPS. Therefore, the conventional SLAM technology has become a bottleneck restricting mobile robot applications. How to develop a new intelligent 3D SLAM technology with characteristics of extremely low power consumption, high efficiency, high robustness and high-level intelligence has become a major problem that need to be overcome urgently. However, in the nature, humans and animals possess superlative navigation capabilities, robustly navigating over large, 3D environments, leveraging an internal neural representation of 3D space combined with external multi-sensory cues and 3D self-motion cues. For example, bats can easily navigate in complex, dynamic, 3D environments efficiently and robustly without acquiring high-precision maps only through the senses of the eyes, ears, and small brains. So how does the brain perform 3D navigation intelligently? In recent years, neuroscientists have gradually discovered internal "3D maps" and "3D compasses" in the brain, which are composed of 3D place cells, 3D head-direction cells, and 3D grid cells, etc., and gradually revealed the mysteries of 3D navigation in the brain. Therefore, can we learn from the 3D navigational neural mechanisms of the biological brain with characteristics of intelligence, efficiency, and robustness to overcome the major issues of intelligent 3D SLAM for mobile robots? This dissertation presents a novel brain-inspired SLAM system for 3D environments named NeuroSLAM, based upon computational models of 3D place cells, 3D grid cells, 3D head direction cells, integrated with a 3D vision perception that provides external visual cues and 3D self-motion cues. The conjunctive 3D place cell and 3D head direction cell models can represent 4DoF (degrees of freedom), 5DoF or 6DoF pose according to the robot’s motion modes adaptively. The conjunctive 3D grid cell models can perform 3D path integration based on 3D visual odometry and perform 3D pose calibration based on local view cues. The 3D grid cell network dynamics drives to perform 3D path integration including 3D translational integration and 3D rotational integration in 3D manifold space. The neural network activities of local view cells, 3D place cells and 3D head direction cells drive the creation of a 3D graphical spatial experience map in a real-time, enabling relocalization and loop closure through sequences of familiar local visual cues. A 3D experience map relaxation algorithm is also used to correct cumulative errors in 3D path integration after loop closure. Using both synthetic and real-world datasets comprising complex, indoor and outdoor 3D environments, the results demonstrate NeuroSLAM consistently producing topologically correct three-dimensional maps which can efficiently support intelligent and autonomous navigation for mobile robots in 3D environments. The main contributions of this dissertation are as follows: Firstly, a set of computational models consisting of 3D place cells, 3D head direction cells is proposed for representing 3D pose. The computational models can represent and maintain 4DoF, 5DoF or 6DoF pose according to the motion mode of the robot adaptively. Secondly, a set of 3D grid cell models is proposed for performing 3D path integration based on 3D visual odometry and performing 3D pose calibration based on local sensory cues. The neural network dynamics drives to perform 3D path integration including 3D translational integration and 3D rotational integration in 3D manifold space, and perform 3D pose calibration based on local sensory cues. This novel approach to 3D path integration can support 3D state estimation using a non-linear form in unknown, unstructured and unpredictable environments with non-Gaussian noise. Thirdly, a novel encoding method of 3D graphical experience map is designed by combining local view cells, 3D place cells, 3D head direction cells and 3D visual odometry, which can not only encode a rich spatial-view place experiences, but also maintain some biological plausibility. This encoding method can also reduce false positives and repeatedly correct loop closure even when facing accumulative odometry error. Fourthly, a light-weight 3D vision perception module is designed for estimating 3D self-motion and local sensory cues via low-cost camera sensors. Using a set of consecutive images with very low resolution, this approach can estimate 3D self-motion and sensory cues efficiently and robustly, which can also overcome the problem of environment change, such as light change, blur, etc. Finally, an open source software system of brain-inspired 3D SLAM is developed. The system is designed based on some advanced software design principles and patterns, such as object-oriented approach, single responsibility principle, interface segregation principle, dependency inversion principle, stable dependencies principle, etc., which is capable of reliability, maintainability, scalability, extensibility, reusability. Keywords: Brain-inspired 3D SLAM, 3D pose representation, 3D path integration, 3D experience map, 3D vision perception, 3D place cells, 3D head-direction cells, 3D grid cells, multi-dimensional continuous attractor network
Article
In this paper, a new mobile robot mapping algorithm inspired from the functionality of hippocampus cells is presented. Place cells in hippocampus can store a map of the environment. This model fuses odometry and vision data based on dimensionality reduction technique, hierarchically. These two types of data are first fused and then considered as inputs to the place cell model. Place cells do the clustering of places. The proposed Place cell model has two types of inputs: Grid cells input and input from the lateral entorhinal cortex (LEC). The LEC is modelled based on the dimension reduction technique. Therefore, the data that causes locations different to be inserted into the place cell from this layer. Another contribution is proposing a new unsupervised dimension reduction method based on k-means. The method can find perpendicular independent dimensions. Also, the distance of cluster centres found in these dimensions is maximised. The method was compared with LDA and PCA in standard functions. Although LDA is a supervised method, the result showed that the proposed unsupervised method outperformed. To evaluate the place cells model, sequences of images collected by a mobile robot was used and similar results to real place cells achieved.
Article
Full-text available
In this paper, we describe a representation for spatial information, called the stochastic map, and associated procedures for building it, reading information from it, and revising it incrementally as new information is obtained. The map contains the estimates of relationships among objects in the map, and their uncertainties, given all the available information. The procedures provide a general solution to the problem of estimating uncertain relative spatial relationships. The estimates are probabilistic in nature, an advance over the previous, very conservative, worst-case approaches to the problem. Finally, the procedures are developed in the context of state-estimation and filtering theory, which provides a solid basis for numerous extensions.
Article
Head direction (HD) cells, abundant in the rat postsubiculum and anterior thalamic nuclei, fire maximally when the rat's head is facing a particular direction. The activity of a population of these cells forms a distributed representation of the animal's current heading. We describe a neural network model that creates a stable, distributed representation of head direction and updates that representation in response to angular velocity information. In contrast to earlier models, our model of the head direction system accurately tracks a series of actual rat head rotations, and, using biologically plausible neurons, it fits the single-cell tuning curves of real HD cells recorded from rats executing those same rotations. The model makes neurophysiological predictions that can be tested using current technologies.