Brain-Based Robot Navigation

Michael Milford

 

The core Theme 3 research topic of brain-based robot navigation has produced significant research outcomes as well as driving collaborative research with computational neuroscientists. Major robotic research outcomes include breakthrough results in visual SLAM (Simultaneous Localization And Mapping) in very large environments as well as in long term persistent robot navigation, both achieved using the brain-based RatSLAM robot navigation system. Major neuroscience research outcomes include the development and publication of a new theory of how grid cells in the rodent brain may help with navigation in perceptually ambiguous environments. Theme research has driven collaborative work with computational neuroscientists working on functional neural learning models, spiking head-direction models implemented on robots, and with roboticists developing a new robotic iRat platform. The work has resulted in two significant funding outcomes which will result in continuation of the further research based off the significant research outcomes from the Thinking Systems Project. Here we present three of the major research outcomes of the theme.

Solving Navigational Uncertainty using Grid Cells on Robots

To successfully navigate their habitats, many mammals use a combination of two mechanisms, path integration and calibration using landmarks, which together enable them to estimate their location and orientation, or pose. In large natural environments, both these mechanisms are characterized by uncertainty: the path integration process is subject to the accumulation of error, while landmark calibration is limited by perceptual ambiguity. It remains unclear how animals form coherent spatial representations in the presence of such uncertainty. Navigation research using robots has determined that uncertainty can be effectively addressed by maintaining multiple probabilistic estimates of a robot’s pose. We showed how conjunctive grid cells in dorsocaudal medial entorhinal cortex (dMEC) may maintain multiple estimates of pose using a brain-based robot navigation system known as RatSLAM. Based both on rodent spatially-responsive cells and functional engineering principles, the cells at the core of the RatSLAM computational model have similar characteristics to rodent grid cells, which we demonstrate by replicating the seminal Moser experiments. We applied the RatSLAM model to a new experimental paradigm designed to examine the responses of a robot or animal in the presence of perceptual ambiguity. Our computational approach enabled us to observe short-term population coding of multiple location hypotheses, a phenomenon which would not be easily observable in rodent recordings. We gathered behavioral and neural evidence demonstrating that the conjunctive grid cells maintain and propagate multiple estimates of pose, enabling the correct pose estimate to be resolved over time even without uniquely identifying cues. While recent research has focused on the grid-like firing characteristics, accuracy and representational capacity of grid cells, our results identify a possible critical and unique role for conjunctive grid cells in filtering sensory uncertainty. We anticipate our study to be a starting point for animal experiments that test navigation in perceptually ambiguous environments.

The actual robot pose and corresponding location estimates encoded by the ensemble grid cell firing. Each firing plot corresponds to various times after the robot was replaced at corner C facing D. (a) Schematic of the robot’s pose corresponding to each of the four ensemble firing plots. (b) Location estimates as encoded by the weighted sum of the firing fields of all active cells at various times. The circle shows the robot’s actual location. Cell firing initially (t = 2.5, 7.2 s) supported and maintained two approximately equal location estimates (1 and 2) – sighting the first black cue did not provide sufficient information to disambiguate the robot’s location. After sighting of the second black cue (t = 12.4 s), cell firing resolved to code primarily for the correct location 1 – location estimate 2 disappeared (t = 15.9 s) and there was limited firing for a new location estimate 3.

Persistent Navigation and Mapping using a Biologically Inspired SLAM System

The challenge of persistent navigation and mapping is to develop an autonomous robot system that can simultaneously localize, map and navigate over the lifetime of the robot with little or no human intervention. Most solutions to the SLAM problem aim to produce highly accurate maps of areas that are assumed static. In contrast, solutions for persistent navigation and mapping must produce reliable goal-directed navigation outcomes in an environment that is assumed to be in constant flux. We investigated the persistent navigation and mapping problem in the context of an autonomous robot that performs mock deliveries in a working office environment over a two week period. The solution was based on the biologically inspired visual SLAM system, RatSLAM. RatSLAM performed SLAM continuously while interacting with global and local navigation systems, and a task selection module that selected between exploration, delivery, and recharging modes. The robot performed 1143 delivery tasks to 11 different locations with only one delivery failure (from which it recovered), travelled a total distance of more than 40 kilometers over 37 hours of active operation, and recharged autonomously a total of 23 times.

A photo of the robot in the environment during the experiments, and a schematic showing the local navigation process. (b) Laser and sonar scans (idealized and sparser than in reality) are converted into a (c) particle obstacle map. Branches from a tree search are grouped and converted into candidate path arcs (d) which are compared for closeness to either a specified local goal or path.

Mapping a Suburb with a Single Camera using a Biologically Inspired SLAM System

This work developed a biologically inspired approach to vision-only SLAM on ground-based platforms. The core SLAM system, dubbed RatSLAM, is based on computational models of the rodent hippocampus, and is coupled with a lightweight vision system that provides odometry and appearance information. RatSLAM builds a map in an online manner, driving loop closure and re-localization through sequences of familiar visual scenes. Visual ambiguity is managed by maintaining multiple competing vehicle pose estimates, while cumulative errors in odometry are corrected after loop closure by a map correction algorithm. We demonstrated the mapping performance of the system on a 66 kilometer car journey through a complex suburban road network. Using only a web camera operating at 10 Hz, RatSLAM generated a coherent map of the entire environment at real-time speed, correctly closing more than 51 loops of up to 5 kilometers in length.

RatSLAM created an accurate street map of the entire suburb of St Lucia (66 km of driving) using only the sensory input from a web camera mounted on top of the car.

Ongoing Collaborative Research

Calibrating Spiking Head-Direction Networks on Robots
Stratton, P., Milford, M.J., Wyeth, G.F., Wiles, J.
Combining spiking head-direction networks, theoretical models of head-direction cells and experiments on robots to provide insights into biology. Calibrating spiking head direction networks on robots with long-term deployments, for example in factory and warehouse delivery tasks, where on-going calibration is required due to mechanical wear and damage accrued over long timeframes. Involves the University of Queensland and the Queensland University of Technology. 2 publications to date.

Blind Bayes in a Box: Rat and Robot Navigation in the Dark
Cheung, A., Ball, D., Milford, M.J., Wyeth, G.F., Wiles, J.
Developing algorithms for calibrating navigation in sensory deprived environments such as in darkness through experimentation on a range of robotic platforms. Involves the University of Queensland and the Queensland University of Technology. 1 paper under review.

A Navigating Rat Animat
Ball, D., Heath, S., Milford, M.J., Wyeth, G.F., Wiles, J.
Development of algorithms to provide mapping and navigation capabilities for a robotic iRat. Development of adaptive exploration and navigation behaviours for a robotic iRat. Involves the University of Queensland and the Queensland University of Technology. 1 publication.

A Flying Rat-Brained Robot
Milford, M.J., Schill, F., Corke, P., Mahony, R., Wyeth, G.F.
Developing a visual SLAM system for mapping and navigation on micro aerial vehicles flying in cluttered indoor and outdoor environments. Involves the Queensland University of Technology and the Australian National University. 1 publication.

Refereed Publications 2007 – 2012

Books

  • Milford, M.J. (2008) Robot Navigation From Nature, Springer Tracts in Advanced Robotics, Volume 41, Springer-Verlag.

Book Chapters

  • Wyeth, G.F., Milford, M.J., Schulz, R., Wiles, J. (2011) The RatSLAM project: robot spatial navigation, in Neuromorphic and Brain-Based Robotics, Cambridge University Press. (PDF File 102 KB)
  • Milford, M.J., Wyeth, G.F. (2007) Spatial Mapping and Map Exploitation: a Bio-Inspired Engineering Perspective, Springer Lecture Notes in Computer Science (LNCS). (PDF File 2,402 KB)

Refereed Journal Articles

  • Maddern, W, Milford, M.J., Wyeth, G.F. (2012 in press), Probabilistic Localisation and Mapping using a Continuous Appearance-based Trajectory, International Journal of Robotics Research.
  • Stratton, P., Milford, M.J., Wyeth, G.F, Wiles, J. (2011) Using Strategic Movement to Calibrate a Neural Compass: A Spiking Network for Tracking Head Direction in Rats and Robots, PLoS One, 6(10).
  • Nolan, C.R., Wyeth, G.F., Milford, M.J., Wiles, J. (2010) The Race to Learn: Spike Timing and STDP Can Coordinate Learning and Recall in CA3. Hippocampus. DOI: 10.1002/hipo.20777 (PDF File 980 KB)
  • Milford, M.J., Wyeth, G.F. (2010) Persistent Navigation and Mapping using a Biologically Inspired SLAM System. The International Journal of Robotics Research 29: 1131. (PDF File 2,194 KB)
  • Milford, M.J., Wiles, J., Wyeth, G.F.(2010) Solving Navigational Uncertainty using Grid Cells on Robots, PLoS Computational Biology, 6 (11), November.   (PDF File 867 KB)
  • Milford, M.J., Wyeth, G.F. (2010) Hybrid Robot Control and SLAM for Persistent Navigation and Mapping. Robotics and Autonomous Systems. Vol. 58, Issue 9, 1096-1104.   (PDF File 1,781 KB)
  • Wyeth, G.F., Milford, M.J. (2009) Spatial Cognition for Robots: Robot Navigation from Biological Inspiration. IEEE Robotics and Automation Magazine, 16 (3). (PDF File 3,948 KB)
  • Milford, M.J., Wyeth, G.F. (2008) Mapping a Suburb with a Single Camera using a Biologically Inspired SLAM System. IEEE Transactions on Robotics, 24 (5), pp. 1038-1053.   (PDF File 1,902 KB)
  • Milford, M.J. (2008) From Rats to Robots: Engineering a Functional Biomimetic Navigation System. Journal and Proceedings of The Royal Society of New South Wales, 141 (1-2), pp. 3-25.
  • Milford, M.J., Schulz, R., Prasser, D., Wyeth, G.F., Wiles, J. (2007) Learning Spatial Concepts from RatSLAM Representations. Robotics and Autonomous Systems, pp. 403-410.   (PDF File 928 KB)

Refereed Conference Papers and Book Chapters

  • Milford, M.J., George, A. (2012) Featureless Visual Processing for SLAM in Changing Outdoor Environments, in proceedings of the International Conference on Field and Service Robotics, Matsushima, Japan.
  • Milford, M.J., Wyeth, G.F. (2012) SeqSLAM: Visual Route-Based Navigation for Sunny Summer Days and Stormy Winter Nights, in proceedings of the IEEE International Conference on Robotics and Automation.
  • Glover, A.J., Maddern, W. P., Warren, M., Reid, S., Milford, M.J., Wyeth, G.F. (2012) OpenFABMAP: An Open Source Toolbox for Appearance-based Loop Closure Detection, in proceedings of the IEEE International Conference on Robotics and Automation.
  • Maddern, W. P., Milford, M.J., Wyeth, G.F. (2012) Capping Computation Time and Storage Requirements for Appearance-based Localization with CAT−SLAM, in proceedings of the IEEE International Conference on Robotics and Automation.
  • Milford, M.J. (2011) Towards Condition-Invariant Sequence-Based Route Recognition, in proceedings of the Australasian Conference on Robotics and Automation, Melbourne.
  • Milford, M.J., McKinnon, D., Warren, M., Wyeth, G.F., Upcroft, B. (2011) Feature-based Visual Odometry and Featureless Place Recognition for SLAM in 2.5D Environments, in proceedings of the Australasian Conference on Robotics and Automation, Melbourne.
  • Milford, M.J., Schill, F., Corke, P., Mahony, R., Wyeth, G.F. (2011) Aerial SLAM with a Single Camera using Visual Expectation, in proceedings of the IEEE International Conference on Robotics and Automation, Shanghai.  (PDF File 1,643 KB)
  • Maddern, W. P., Milford, M.J., Wyeth, G.F. (2011)  Continuous Appearance-based Trajectory SLAM, in proceedings of the IEEE International Conference on Robotics and Automation, Shanghai. (PDF File 302 KB)
  • Schulz, R., Glover, A., Milford, M.J., Wyeth, G.F., Wiles, J. (2011) Lingodroids: Studies in Spatial Cognition and Language, in proceedings of the IEEE International Conference on Robotics and Automation, Shanghai. (PDF File 691 KB)
  • Milford, M.J., Maddern, W., Wyeth, G.F. (2011) A Two Week Persistent Navigation and Mapping Experiment using RatSLAM: Insights and Current Developments, in proceedings of the International Conference on Robotics and Automation Workshop on Long-term Autonomy, Shanghai.
  • Ball, D., Heath, S., Milford, M.J., Wyeth, G.F., Wiles, J. (2010) A Navigating Rat Animat, in proceedings of the International Conference on Artificial Life, Odense, Denmark. (PDF File 1,010 KB)
  • Glover, A.J., Maddern, W. P., Milford, M.J., Wyeth, G.F. (2010) FAB-MAP + RATSLAM: Appearance-based SLAM for Multiple Times of Day. In proceedings of the IEEE International Conference on Robotics and Automation, Anchorage, Alaska. (PDF File 493 KB)
  • Milford, M.J., Wyeth, G.F. (2010) Improving Recall in Appearance-Based Visual SLAM using Visual Expectation, Australasian Conference on Robotics and Automation, Brisbane, Australia. (PDF File 857 KB)
  • Maddern, W., Milford, M.J., Wyeth, G.F. (2010) Loop Closure Detection on a Suburban Road Network using a Continuous Appearance-based Trajectory, in proceedings of the Australasian Conference on Robotics and Automation, Brisbane.
  • Nolan, C.R., Wyeth, G.F., Milford, M.J., Wiles, J. (2010) A neural microcircuit using spike timing for novelty detection, in Computational and Systems Neuroscience, Salt Lake City, United States.
  • Wyeth, G.F., Milford, M.J. (2009) Towards Lifelong Navigation and Mapping in an Office Environment, in proceedings of the International Symposium of Robotics Research.
  • Stratton, P., Milford, M.J., Wiles, J., Wyeth, G.F. (2009) Automatic Calibration of a Spiking Head-Direction Network for Representing Robot Orientation, in proceedings of the Australasian Conference on Robotics and Automation, Sydney, Australia. (PDF File 2,071 KB)
  • Maddern, W. P., Glover, A.J., Milford, M.J., Wyeth, G.F. (2009) Augmenting RatSLAM using FAB-MAP-based Visual Data Association. In proceedings of the Australasian Conference on Robotics and Automation, Sydney, Australia. (PDF File 1,284 KB)
  • Oates, R., Milford, M.J., Wyeth, G.F., Kendall, G., Garibaldi, J. M. (2009) The Implementation of a Novel, Bio-Inspired, Robotic Security System. In proceedings of the IEEE International Conference on Robotics and Automation, Kobe, Japan. (PDF File 687 KB)
  • Milford, M.J., Wyeth, G.F. (2008) Single Camera Vision-Only SLAM on a Suburban Road Network, in proceedings of the IEEE International Conference on Robotics and Automation, Pasadena, United States.   (PDF File 1,535 KB) 
  • Milford, M.J., Wyeth, G.F. (2007) Featureless Vehicle-Based SLAM with a Consumer Camera, in proceedings of the Australasian Conference on Robotics and Automation, Brisbane, Australia.    (PDF File 1,029 KB)

Submissions Under Review

  • Milford, M.J., George, A. (2012) Featureless Visual Processing for SLAM in Changing Outdoor Environments, in proceedings of the International Conference on Field and Service Robotics, Matsushima, Japan.
  • Milford, M.J., Wyeth, G.F. (2012) SeqSLAM: Visual Route-Based Navigation for Sunny Summer Days and Stormy Winter Nights, in proceedings of the IEEE International Conference on Robotics and Automation.
  • Glover, A.J., Maddern, W. P., Warren, M., Reid, S., Milford, M.J., Wyeth, G.F. (2012) OpenFABMAP: An Open Source Toolbox for Appearance-based Loop Closure Detection, in proceedings of the IEEE International Conference on Robotics and Automation.
  • Maddern, W. P., Milford, M.J., Wyeth, G.F. (2012) Capping Computation Time and Storage Requirements for Appearance-based Localization with CAT−SLAM, in proceedings of the IEEE International Conference on Robotics and Automation.

Patents / Commercialisation

Milford, M.J., Wyeth, G.F. (2009) “Method and Algorithm for Creating a Navigable Spatial Map of Panoramic Images from Video”, filed provisionally.

Related Activities

RSS Program Committee 2011, 2012
Associate Editor for International Conference on Robotics and Automation 2012
Australasian Conference on Robotics and Automation Program Committee 2010, 2011
Australasian Conference on Robotics and Automation Local Chair 2010

Invited Talks

  • Invited speaker at the Centre for Memory and Brain at Boston University (Boston, United States, 2012)
  • Invited speaker at the Workshop on Long-term Autonomy at the International Conference on Robotics and Automation (St. Paul, United States, 2012)
  • Invited speaker at the Office of Naval Research Multidisciplinary University Initiative at Boston University (Boston, United States, 2012)
  • Invited speaker at the Workshop on Long-term Autonomy at the International Conference on Robotics and Automation (Shanghai, China, 2011)
  • Invited speaker at the Workshop on Bio-mimetic and Hybrid Approaches to Robotics at the International Conference on Robotics and Automation (Shanghai, China, 2011)
  • Invited presentation at the Champalimaud Neuroscience Hippocampus and Navigation Symposium (Lison, Portugal, 2008)
  • Invited presentation at the Kavli Institute to Professors Edvard and May-Britt Moser, the discoverers of grid cells (Trondheim, Norway, 2008).
  • Invited presentation at the internationally renowned laboratory of Professor John O’Keefe, the discover of place cells, and Professor Neil Burgess, the discoverer of boundary vector cells (London, United Kingdom, 2008).
  • Invited presentation at the Australian National University in the Computer Vision and Robotics Lecture Series (Canberra, Australia, 2010)
  • Invited presentation at the Computational Neuroscience course run at the Queensland Brain Institute, rated one of the top talks by the audience (Brisbane, Australia, 2009)
  • More than a dozen school and community presentations on robotics and neuroscience (Australia)

International and National Collaboration

  • Visit from Dr Justin Werfel, Wyss Institute for Biologically Inspired Engineering, Harvard University, Dec 2011.
  • Visits from Associate Professor Robert Mahony, Faculty of Engineering and Information Technology, Australian National University, Canberra, Australia. February and October 2010.
  • Visit to the laboratory of Associate Professor Robert Mahony, Faculty of Engineering and Information Technology, Australian National University, Canberra, Australia. February 2010.
  • Visit to the laboratory of Professor Edvard Moser and Professor May-Britt Moser, Kavli Institute for Systems Neuroscience and Centre for Biology for Memory, Norwegian University of Science and Technology, Trondheim, Norway, March 2008.
  • Visit to laboratories of Professor Neil Burgess and Professor John O’Keefe, University College London Institute of Cognitive Neuroscience, London, United Kingdom, March 2008.
  • Visit to the laboratory of Dr Andrew Davison, Department of Computing, Imperial College London, March 2008
  • Visit to the laboratory of Professor Keith Downing, Department of Computer and Information Sciences, Norwegian University of Science and Technology, March 2008.
  • Visit from Dr Robert Oates, School of Computer Science and IT, University of Nottingham, Nottingham, United Kingdom, 2008.

Media Coverage

Supervision

2012 onwards – Primary Supervisor, PhD Candidate Zetao Chen
2011 onwards – Associate Supervisor, PhD Candidate Stephanie Reid
2009 onwards – Associate Supervisor, PhD Candidate William Maddern

2012 – Undergraduate Supervisor, Thesis Student Adam Jacobson
2011 – Undergraduate Supervisor, Summer Student Ashleigh George
2008 – Undergraduate Supervisor, Thesis Student Jeremy Norton
2007 – 2008 – Supervisor, Research Assistant William Maddern

Positions after Thinking Systems

I was appointed as a permanent Lecturer in July 2011 to the Queensland University of Technology. In 2012 I commenced a three year Discovery Early Career Researcher Award Fellowship at the same university conducting research on Visual navigation for sunny summer days and stormy winter nights. I am also be leading a Discovery Project Grant on Brain-based Sensor Fusion for Navigating Robots (see Grants for details).

 

Continuous Appearance-based Localisation and Mapping

Will Maddern

The future capabilities of mobile robots depend strongly on their abilities to navigate and interact in the real world. A key requirement for navigation is an internal representation of the environment that the robot inhabits, along with the robot's location within the environment. Determining this using only information from onboard sensors is commonly referred to as Simultaneous Localisation and Mapping (SLAM). To avoid computational and scaling limitations, a number of SLAM approaches forsake geometric accuracy for flexibility to form semi-metric or non-metric 'topological' approaches. The two most successful non-geometric SLAM algorithms are RatSLAM and FAB-MAP. RatSLAM is based on a computational model of the rodent hippocampus, and is not a probabilistic system. FAB-MAP forsakes map building entirely and instead focuses on visual data association (so-called 'SLAM in appearance space').

My project aims to develop a novel SLAM solution grounded in probabilistic SLAM techniques, but incorporating characteristics of both RatSLAM and FAB-MAP to produce a highly capable, generally applicable solution for mobile robotics. The system will use vision as the primary sensory method, and generate a novel semi-metric trajectory-based map. Additionally, the system will require minimal tuning or calibration to work in a particular environment to provide greater ease of use; as with probabilistic SLAM systems the majority of parameters will be measurable properties of the robot platform. The key innovations of this system will be the use of continuous, non-geometric feature representation and trajectory-based pose filtering, both characteristics of RatSLAM, as well as principled visual data association in the style of FAB-MAP.

Published Work

Published work arose from both an initial study into a combination of FAB-MAP and RatSLAM. We performed two large-scale outdoor experiments with a hybrid RatSLAM-FAB-MAP system, demonstrating that it could successfully map a 66km route through an entire suburb as well as an 18km route repeated at many different times of day across a period of 3 weeks. This study served to illustrate many of the shortcomings of both algorithms; neither algorithm alone could complete the full day mapping, and many of the inaccuracies were due to the underlying feature detection rather than the probabilistic data association.

Current Work

I developed a new approach to probabilistic visual matching combined with vehicle odometry. This approach combines optical flow on a spherical plane with a bag-of-words matching algorithm to extrapolate the appearance of a visual scene from an initial location plus an arbitrary displacement. By performing this repeatedly over multiple displacements with multiple estimates of the average distance to features, the mean distance can be found using ML estimation. The study demonstrated that extrapolating feature appearance over a distance of 20m increased the matching performance by an average of 12dB over static image matching (as used by FAB-MAP). Figure 1 illustrates the feature extrapolation process. I am currently extending the feature extrapolation process to include interpolation between multiple images, with the goal of forming a continuous feature representation from a series of discrete snapshots of an environment with accompanying visual odometry.

RatSLAM-created experience maps

Figure 1. RatSLAM-created experience maps and loop closure locations using FAB-MAP data association. Sequentially including (a) 8:45am (b) 10:00am (c) 12:10pm (d) 2:10pm and (e) 3:45pm.

RatSLAM-created experience maps

Figure 2. Predicted feature locations and uncertainty for a sample sequence of images. Feature locations are denoted by green circles and the Monte Carlo sampled locations for each feature are shown as red dots approximating the feature covariance.

I developed a new approach to probabilistic pose filtering along a trajectory using techniques from FastSLAM and FAB-MAP. The novel algorithm, dubbed Continuous Appearance-based Trajectory SLAM (CAT-SLAM), conditions the joint distribution of the observation and motion model on a continuous trajectory of previously visited locations. The distribution is evaluated using a Rao-Blackwellised particle filter, which represents location hypotheses as particles constrained to the trajectory. We compared the performance of CAT-SLAM to FAB-MAP (an appearance-only SLAM algorithm) in an outdoor environment, demonstrating a threefold increase in the number of correct loop closures detected by CAT-SLAM, illustrated in Figure 3. The results of the mapping experiment demonstrated that the combination of both appearance and motion information in CAT-SLAM provides a clear advantage over appearance-based SLAM systems that rely on visual data alone for applications that require 100% precision loop closure. The improvement over FAB-MAP is twofold; first, the addition of a pose filter allows spurious false positives to be rejected, and it allows a location hypothesis to be maintained with only partial visual matches.

Loop closures projected on GPS ground truth

Figure 3. Loop closures projected on GPS ground truth for a) FAB-MAP and b) CAT-SLAM. Lighter green points indicate true positives and darker blue points false negatives. Insets illustrate sequential loop closure performance.

Conference Abstracts

  • Maddern, W. P.,  Milford, M.J., Wyeth, G.F. (2011)  “Continuous Appearance-based Trajectory SLAM”, in proceedings of the IEEE International Conference on Robotics and Automation, Shanghai. (PDF File 302 KB)
  • Maddern, W. P., Glover, A.J., Milford, M.J., Wyeth, G.F. (2009) “Augmenting RatSLAM using FAB-MAP-based Visual Data Association”, in proceedings of the Australasian Conference on Robotics and Automation, Sydney, Australia. (PDF File 1,284 KB)
  • Glover, A., Maddern, W., Milford, M.J., Wyeth, G.F. (2010) “FABMAP + RATSLAM: Appearance-based SLAM for Multiple Times of Day”, IEEE International Conference on Robotics and Automation 2010, Anchorage, Alaska.   (PDF File 493 KB)