SLAM Constructor Framework

The SLAM constructor framework provides common functionality and classes that may be used to create custom SLAM algorithms (currently only 2D laser scan-based methods are supported). It also includes implementation of several SLAM algorithms: tinySLAM, vinySLAM and GMapping. They show an example of the framework usage and can be used as a base of a new SLAM algorithm.

Hardware Requirements

Current implementation requires odometry data and laser scans to be provided by the ROS topics (see Subscribed topics). It also supposes that a laser scanner is fixed in (0, 0) of a robot and mounted horizontally.

Examples

Provided algorithms usage

Each algorithm is supplied with launch-files for the MIT Stata Center and PR2 – Willow Garage datasets that give the idea of how to launch algorithms. The _mit_ launch-files can be used in general case if data provided by a dataset do not require any preprocessing. For example, tinySLAM can be launched in the following way:

roslaunch slam_constructor tiny_mit_run.launch path:=[path to dataset]

In order to run an algorithm on data received in real time you can remove a dataset player node from a launch file, but make sure that sensor data are provided through subscribed topics.

Single-hypothesis SLAM workflow

The following diagram shows the expected workflow of the general single-hypothesis SLAM algorithm:

Parameters

The following parameters are common for all provided SLAM algorithms:

ROS-specific parameters

• ~in/lscan2D/ros/topic/name (string, default: /base_scan) – the laser scan topic name
• ~in/odometry/ros/tf/odom_frame_id (string, default: odom_combined) – the odometry tf frame id
• ~ros/tf/map_frame_id (string, default: map) – the map tf frame id
• ~ros/tf/robot_pose_frame_id (string, default: robot_pose) – the output robot pose frame id
• ~ros/tf/async_correction (bool, default: false) – set to true to publish the map → odom transformation asynchronous (in a different thread)
• tf listener parameters (refer to tf::MessageFilter documentation):
  • ~ros/subscribers_queue_size (int, default: 1000) – the queue size for message_filters::Subscriber
  • ~ros/tf/buffer_duration (double, default: 5.0) – the buffer size in seconds for tf::TranformListner
  • ~ros/filter_queue_size (int, default: 1000) – the filter queue size
• ~ros/rviz/map_publishing_rate (double, default: 5.0) – the map publishing interval in seconds
• ~ros/skip_exceeding_lsr_vals (bool, default: false) – set to true to skip laser scan outliers or to false to insert such measurements in a map as a free space

General SLAM parameters

• ~slam/map/height_in_meters (double, default: 10.0) – the map height in meters
• ~slam/map/width_in_meters (double, default: 10.0) – the map width in meters
• ~slam/map/meters_per_cell (double, default: 0.1) – the map resolution in meters
• ~slam/performance/use_trig_cache (bool, default: false) – use trigonometry cache to speed up scan point operations
• cell occupancy estimator parameters:
  • ~slam/occupancy_estimator/type (string, default: const) – the type of the cell occupancy estimator:
    • const – a cell is considered occupied if a laser scan point is inside the cell
    • area – a cell occupancy estimation is based on how a laser beam splits the cell
  • ~slam/occupancy_estimator/base_occupied/prob (double, default: 0.95) – the cell’s probability to be occupied if a scan point is inside the cell
  • ~slam/occupancy_estimator/base_occupied/qual (double, default: 1.0)
  • ~slam/occupancy_estimator/base_empty/prob (double, default: 0.01) – the cell’s probability to be empty if a laser beam passes through the cell
  • ~slam/occupancy_estimator/base_empty/qual (double, default: 1.0)
• ~slam/mapping/blur (double, default: 0.0) – blur obstacles along the direction of a laser beam in the specified range (in meters)
• ~slam/mapping/max_range (double, default: <infinity>) – maximum valid range for laser scan measurements when updating a map with a new laser scan

Scan-matcher parameters

• ~slam/scmtch/type (string, default: <undefined>) – the scan matcher (SM) to be used. Currently the following scan matchers are supported:
  • MC – Monte-Carlo scan matcher. Parameters:
    • ~slam/scmtch/MC/attempts_limit (unsigned int, default: 100) – the maximum number of SM iterations
    • ~slam/scmtch/MC/seed (int, default: <random>) – the seed value for RNG
    • ~slam/scmtch/MC/dispersion/translation (double, default: 0.2)
    • ~slam/scmtch/MC/dispersion/rotation (double, default: 0.1)
    • ~slam/scmtch/MC/dispersion/failed_attempts_limit (unsigned int, default: 20)
  • HC – hill climbing scan matcher. Parameters:
    • ~slam/scmtch/HC/distortion/translation (double, default: 0.1) – the initial step size in x/y direction in meters
    • ~slam/scmtch/HC/distortion/rotation (double, default: 0.1) – the initial step size in th direction in radians
    • ~slam/scmtch/HC/distortion/failed_attempts_limit (unsigned int, default: 6)
  • BF – brute-force scan matcher. Searches for the best scan position around the initial guess. Parameters set the search ranges and search steps for each coordinate (in meters for translation, radians for rotation):
    • ~slam/scmtch/BF/[x, y, t]/from (double, default: -0.5, -0.5, -5°)
    • ~slam/scmtch/BF/[x, y, t]/to (double, default: 0.5, 0.5, 5°)
    • ~slam/scmtch/BF/[x, y, t]/step (double, default: 0.1, 0.1, 1°)
• ~slam/scmtch/spe/type (string, default: <undefined>) – the scan probability estimator type. Currently only wmpp (weighted mean point probability) is supported. Parameters:
  • ~slam/scmtch/spe/wmpp/weighting/type (string, default: <undefined>)
    • even – each point in a scan has the equal weight
    • viny – weighting scheme used in vinySLAM (see the paper)
    • ahr – weighting scheme based on angle histograms
  • ~slam/scmtch/spe/wmpp/sp_skip_rate (unsigned int, default: 0) – skip every n-th point in scan
  • ~slam/scmtch/spe/wmpp/sp_max_usable_range (double, default: -1.0) – max valid scan measurement range used in scan probability estimation
• ~slam/scmtch/oope/type (string, default: obstacle) – the occupancy observation probability estimator type. Currently the following types are supported:
  • obstacle
  • max
  • mean
  • overlap

tinySLAM parameters

• ~slam/cell/type (string, default: base) – the cell model type. Accepted values:
  • base – the cell model used in original tinySLAM
  • avg – the cell model proposed in this paper

GMapping parameters

GMapping has the following additional parameters (note that ~slam/scmtch/oope/type shouldn’t be provided or must be custom):

• ~slam/particles/number (unsigned int, default: 30)
• ~slam/particles/sample/xy/mean (double, default: 0.0)
• ~slam/particles/sample/xy/sigma (double, default: 0.1)
• ~slam/particles/sample/theta/mean (double, default: 0.0)
• ~slam/particles/sample/theta/sigma (double, default: 0.03)
• ~slam/particles/sm_delta_lim/xy/min (double, default: 0.6)
• ~slam/particles/sm_delta_lim/xy/max (double, default: 0.8)
• ~slam/particles/sm_delta_lim/theta/min (double, default: 0.3)
• ~slam/particles/sm_delta_lim/theta/max (double, default: 0.4)
• ~slam/scmtch/oope/custom/fullness_threshold (double, default: 0.1)
• ~slam/scmtch/oope/custom/window_size (unsigned int, default: 1)

Subscribed topics

• /tf (tf/tfMessage) – should provide required transforms.
• /base_scan (sensor_msgs/LaserScan) – laser scans that should be used to create a map

Published topics

• map (nav_msgs/OccupancyGrid) – the map data are periodically published to this topic

Required tf transforms

• <the frame attached to incoming scans> (extracted from scan messages) → odom_frame_id (configurable, default: odom_combined) - usually provided by the odometry system

Provided tf transforms

• The current estimate of the robot’s pose within the map frame. Frame ids can be configured.
  • map_frame_id → odom_frame_id
  • map_frame_id → robot_pose_frame_id

Offline mode

The package provides the utility lslam2d_bag_runner to launch algorithms in the offline mode to process datasets in BAG format.

Usage

lslam2d_bag_runner <slam type> <bag file>
                   [-v] [-t <traj file>] [-m <map file>]
                   [-p <properties file>]

Parameters

• <slam type> – viny, tiny or gmapping
• <bag file> – the path to a dataset
• -v – enable verbose output
• -t <traj file> – save a robot trajectory to traj file in TUM format
• -m <map file> – save an output map to map file in PGM format
• -p <properties file> – the path to a SLAM configuration file in key=value format. Example configurations can be found here

Contributors

• Artur Huletski
• Dmitriy Kartashov
• Kirill Krinkin

Copyright (c) 2018 JetBrains Research, Mobile Robot Algorithms Laboratory