LaMa - A Localization and Mapping library.

Overview

LaMa is a C++11 software library for robotic localization and mapping developed at the Intelligent Robotics and Systems (IRIS) Laboratory from the University of Aveiro - Portugal. It includes a framework for 3D volumetric grids (for mapping), a localization algorithm based on scan matching and two SLAM solution (an Online SLAM and a Particle Filter SLAM).

The main feature is efficiency. Low computational effort and low memory usage whenever possible. The minimum viable computer to run our localization and SLAM solutions is a Raspberry Pi 3 Model B+.

Build

To build LaMa, clone it from GitHub and use CMake to build.

$ git clone https://github.com/iris-ua/iris_lama
$ cd iris_lama
$ mkdir build
$ cd build
$ cmake ..

Its only dependency is Eigen3. Note: LaMa does not provide any executable. For an example on how to use it, please take a look at our integration with ROS.

Integration with ROS

The source code contains package.xml so that it can be used as a library from external ros packages. We provide ROS nodes to run the localization and the two SLAM solutions. Please go to iris_lama_ros for more information.

Sparse-Dense Mapping (SDM)

Sparse-Dense Mapping (SDM) is a framework for efficient implementation of 3D volumetric grids. Its divides space into small dense patches addressable by a sparse data-structure. To improve memory usage each individual patch can be compressed during live operations using lossless data compression (currently lz4 and Zstandard) with low overhead. It can be a replacement for OctoMap.

Currently it has the following grid maps implemented:

Distance Map: It provides the distance to the closest occupied cells in the map. We provide the DynamicDistanceMap which is an implementation of the dynamic Euclidean map proposed by:

B. Lau, C. Sprunk, and W. Burgard Efficient Grid-Based Spatial Representations for Robot Navigation in Dynamic Environments Robotics and Autonomous Systems, 61 (10), 2013, pp. 1116-1130, Elsevier
Occupancy Map: The most common representation of the environment used in robotics. Three (3) variants of the occupancy map are provided: a SimpleOccupancyMap where each cell has a tri-state: free, occupied or unknown: a ProbabilisticOccupancyMap that encodes the occupancy probability of each cell with logods; and a FrequencyOccupancyMap that tracks the number of times a beam hits or traverses (miss) a cell and calculates a hit/miss ratio.

For more information about SDM please read

Eurico Pedrosa, Artur Pereira, Nuno Lau
A Sparse-Dense Approach for Efficient Grid Mapping
2018 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)

Localization based on Scan Matching

We provide a fast scan matching approach to mobile robot localization supported by a continuous likelihood field. It can be used to provide accurate localization for robots equipped with a laser and a not so good odometry. Nevertheless, a good odometry is always recommended.

Eurico Pedrosa, Artur Pereira, Nuno Lau
Efficient Localization Based on Scan Matching with a Continuous Likelihood Field
2017 IEEE International Conference on Autonomous Robot Systems and Competitions (ICARSC)

Online SLAM

For environments without considerable loops this solution can be accurate and very efficient. It can run in real time even on a low-spec computer (we have it running on a turtlebot with a raspberry pi 3B+). It uses our localization algorithm combined with a dynamic likelihood field to incrementally build an occupancy map.

For more information please read

Eurico Pedrosa, Artur Pereira, Nuno Lau
A Non-Linear Least Squares Approach to SLAM using a Dynamic Likelihood Field
Journal of Intelligent & Robotic Systems 93 (3-4), 519-532

Multi-threaded Particle Filter SLAM

This Particle Filter SLAM is a RBPF SLAM like GMapping and it is the extension of the Online SLAM solution to multiple particles with multi-thread support. Our solution is capable of parallelizing both the localization and mapping processes. It uses a thread-pool to manage the number of working threads.

Even without multi-threading, our solutions is a lightweight competitor against the heavyweight GMapping.