This paper presents a new implementation of NeoSLAM, a biologically inspired SLAM system, and provides a comparative evaluation with RatSLAM. The new NeoSLAM implementation incorporates seven main modules: Visual Feature Extractor, Binary Projector, Neocortex, Spatial-View Cells, Pose Cells, Experience Map and Visual Odometry, all implemented as ROS 2 nodes. Experimental evaluation uses three distinct datasets: the Robotarium dataset (5453 frames at 13 FPS), the iRat Australia dataset (16657 frames at 10 FPS), and the FIU MMC Lake dataset collected from a USV at Florida International University.
System Overview
The new implementation of NeoSLAM comprises seven main modules: Visual Feature Extractor, Binary Projector, Neocortex, Spatial-View Cells, Pose Cells, Experience Map and Visual Odometry. The modules are described as ROS 2 nodes, along with the corresponding ROS topics describing the data flow between them.
Neocortex Node
The Neocortex Node transforms the reduced binary visual information into a format that captures spatial and temporal features grounded in the principles of Hierarchical Temporal Memory (HTM). To facilitate understanding of this module, the fundamental structures required for operating an HTM network are described: Sparse Distributed Representation (SDR), Spatial Pooler (SP), and Temporal Memory (TM).
The overlap between two SDRs measures similarity by counting shared active bits:
The overlap is computed as the number of active bits common to both SDRs
A match occurs when the overlap exceeds a predefined threshold:
A match is declared when the overlap value is greater than or equal to the threshold
Robotarium Dataset
Experimental conditions: input rate = 13 FPS, total input frames = 5453, repetitions = 3, experiment duration = 424.8 (s).
iRat Australia Dataset
The iRat Australia dataset features an iRat robot with a differential wheel drive, a forward-facing camera, and wheel encoders providing odometric readings. A camera mounted overhead captured images that enabled the extraction of ground-truth information.
Experimental conditions: input rate = 10 FPS, total input frames = 16657, repetitions = 3, experiment duration = 15.9 (min).
FIU MMC Lake Dataset
Data were collected using a SeaRobotics Surveyor USV at the Green Library Lake at Florida International University (FIU), Modesto Maidique Campus (MMC). The trajectory followed by the USV covers approximately 900 meters over approximately 16 minutes. The dead reckoning data was also recorded and analyzed.
Frequently Asked Questions
How does the new NeoSLAM implementation differ from the original? The new implementation reorganizes the system into seven ROS 2 nodes with clearly defined topics for data flow, enabling modular operation and easier integration with modern robotics frameworks.
What is the role of the Neocortex module in NeoSLAM? The Neocortex Node transforms binary visual information using Hierarchical Temporal Memory principles, applying Sparse Distributed Representations, a Spatial Pooler, and Temporal Memory to capture spatial and temporal features.
What datasets were used to evaluate the new NeoSLAM? Three datasets were used: the Robotarium dataset (5453 frames at 13 FPS), the iRat Australia dataset (16657 frames at 10 FPS), and the FIU MMC Lake dataset collected from a USV covering 900 meters over 16 minutes.
What hardware platforms are represented in the evaluation datasets? The datasets include a differential wheel drive robot (iRat), an overhead-camera tracked robot (Robotarium), and an autonomous surface vehicle (SeaRobotics Surveyor USV), covering both ground and marine robotic platforms.
Acknowledgments
This research was supported in part by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001, Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), Fundação Carlos Chagas Filho (FAPERJ) under grant E-26/204.669/2024, and NSF grants IIS-2024733 and IIS-2331908, the Office of Naval Research grant N00014-23-1-2789, the U.S. Department of Defense grant 78170-RT-REP, and the Florida Department of Environmental Protection grant INV31.