Movebase with Sparse-Waypoint Goals

This package has been derived from the original move_base node from ROS Navigation package. It implements the move_base_swp node, which extends the interface of the the original move_base node to include the sparse waypoints that the robot must visit along its route to the goal.

‘move_base_swp’ is a drop-in replacement for move_base, designed to work with all existing plugins (e.g. DWA planner, navfn, costmap2d) from the Navigation package. The interface is fully compatible with move_base and will accept single-pose goals and behave just as regular move_base with a couple of improvements listed below. The new interface accepts the sparse-waypoint goals which specifies a list of poses, with last pose being the final goal. If a single-waypoint goal is provided the goal reduces to a final goal with no waypoints only issued over the new interface. All other components will work, so move_base_swp is simply a better move_base.

Feature Overview

In this section, we describe the improvements that move_base_swp implements on the top of original move_base node.

Sparse-waypoint goals

move_base_swp implements a new actionlib interface, which allows the caller to define the goal in form of a list of waypoints. Unlike dense waypoints generated by the global planner, this list is sparse. Typical size of this list will include no more than 5-10 waypoints, sometimes just one. The path between sparse waypoints is still within the responsibility of the global planner. In other words, the input to move_base_swp is a list of sparse waypoints that represent an the goal along with a list of interim checkpoints that the robot must visit along its path. The global planner produces a list of dense waypoints in the same manner as in the original move_base, which are fed to the local planner, which, in turn, produces the velocity vector that creates the motion.

While it is theoretically possible to supply a dense set of waypoints over the interface and effectively override the global planner, this is not intended usage. Such usage will likely result in poor performance because of frequent calls to the global planner. The intended usage is for situations when the robot must get from point A to point B, but the application requires that it visits point C along the way. Simply issuing the goal at point B would result in finding the shortest path within the constraints of the cost map which may or may not include point C. If the point C is on a “detour” path it will not be visited. Using move_base_swp allows the application to specify this additional constraint.

While it is possible to achieve the described behavior by first issuing the goal to point C and then as the robot approaches that goal issue the new goal to point B, this will likely result in unwanted slowdown or even complete stop at the interim waypoint. Further, an application would have to constantly monitor the progress of the first goal and issue the next goal “just in time” to ensure smooth transition. Often the “just in time” would imply using ill-defined heuristics that heavily depend on local and global planner tuning parameters.

move_base_swp solves the problem by first calling the global planner for each segment of the path (A-to-C segment, followed by C-to-B segment in our example) and then concatenating the two plans into one global-planner solution before feeding it to the local planner. The result is a smooth motion at the concatenation point without the need for application to intervene in any way. The application simply follows the progress towards the specified goal which includes the specified sparse waypoints.

Windowed global plan

move_base_swp provides an option to feed the horizon of the global plan that is being fed into the local planner. Instead of feeding an entire global plan, move_base_swp has an option to feed only a specified number of plan waypoints starting from the present location and load additional plan waypoints as the robot progresses towards the goal.

This feature can improve motion behavior in general and can stand on its own regardless of whether sparse-waypoint plan is used or not. However, the problem that the windowed global plan solves is exacerbated by the usage of sparse waypoints.

An example of the problem that exists even in the original move_base package is illustrated in this video:

https://youtu.be/7vnxZ25MzEk

The robot is moving from one side to the other of an X-shaped obstacle. Notice how at 0:57 it enters the state where it is not making progress. This happens because the entire area of interest is within the local planner window (we used DWA planner for this experiment). The local planner has the choice to either produce the velocity vector that takes it along the route prescribed by the global plan (the top path shown in green) or to deviate from the global plan and move the robot down the bottom path. From the local planner’s perspective, both paths have equal cost for the set of parameters it is using and the solver is bouncing back and forth between two options without making much progress.

While one motion may eventually prevail, the time it will take for this to happen, is not bounded. Changing the weight factors in the DWA planner to prefer sticking to the global plan more than advancing towards the target (i.e. cutting corners) would resolve this case, but there is always another case for which the same phenomena would happen. Triggering the patience timeout on the local planner and requesting the replan will in practice often break this live-lock, but it would be preferable if we could avoid the need to timeout. Making the local planner horizon smaller would also help in this case, but at the expense of degraded obstacle avoidance performance.

This video shows how the windowed global plan solves the problem:

https://youtu.be/VPTIrmf1TQQ

The thin green line shows the plan as determined by the global planner, which goes all the way to the goal. The thick green line shows the section of the global plan to which the local planner has been exposed. As the robot makes progress, the move_base_swp feeds additional waypoints to the local planner. The new waypoints can be fed either continuously or when the number of presently used waypoints runs below the low-watermark threshold. Because the local planner is only exposed to limited number of waypoints, effectively cutting its horizon, the secondary viable path that causes the live-lock has been eliminated.

The next video shows how the usage of sparse waypoints can be at odds with the local planner:

https://youtu.be/x_Khtj9rBhc

The robot is in the open space (no obstacles) and it gets a goal with two waypoints that define a zig-zag path to the final pose. If the local planner sees the whole plan, it will eventually start cutting corners as the opportunistic behavior of reducing the distance to the goal starts to prevail. The effect is especially visible for the second waypoint where the robot does not even attempt to visit it, but instead cuts straight to the goal.

This video shows how the windowed global plan solves the problem:

https://youtu.be/8t-57WI_6UY

By exposing the local planner to a shorter subset of the global plan, the robot is forced to head towards the waypoint, before it learns about the existence of the sharp turn and gives the opportunity to the local planner to cut the corner.

Brake rampdown

If the goal is canceled or preempted, or the robot is made stop by any means other than arriving to the goal, the original move_base node would abruptly bring the velocity vector to zero. This can result in a jerky motion, which can cause the stress or even damage to robot’s mechanical or electrical system. move_base_swp introduces the break-rampdown feature which makes the robot slow down gradually. The rampdown rate is controlled by a parameter and the behavior can be turned off, essentially resulting on original move_base behavior.

Handbrake interface

Sometimes, an application may decide to temporarily stop the robot and then after some time allow it to proceed towards its target. In the original move_base the only way to achieve this is to cancel the goal and re-issue the new goal after the application decides to restore the motion. move_base_swp provides a dedicated handbrake interface that implements this feature. When the robot is ordered to handbrake, it will not cancel the goal. The goal and the planner will remain active until the handbrake is released, after which the robot will continue to pursue its original plan.

Code improvements

Besides extended functionality, move_base_swp also refactors and improves the general move_base code. Many instances of repeated code are extracted into common functions. Blocks of code embedded in long functions are extracted into separate functions and named accordingly. Finally, some locks are reworked to improve the interlocking correctness.

Interfaces

move_base_swp provides all interfaces that move_base does unchanged. If the move_base node in an application designed to work with it is replaced with the move_base_swp node, the behavior will remain identical. Other than launching the new mode no change to the application would be necessary. However, to use the new capabilities (described above), the application would have to be extended to use the new interfaces. This makes the transition to move_base_swp easy, because the node can be first swapped out, followed by gradually converting the application to use the new features.

This section describes the interfaces (topics and actionlib interfaces) that allow the application to use the new capabilities of the `move_base_swp’

Sparse-waypoint actionlib interface.

This is a standard actionlib interface. Actions are defined in ‘move_base_swp/action/MoveBaseSWP.action file. The standard topics associated with this interface are as follows, move_base_swp/goal, move_base_swp/result, move_base_swp/status, move_base_swp/feedback, and move_base_swp/cancel`. The goal supplied is the list of poses that represent the waypoints that the robot should visit. The last pose in the list is the final goal:

geometry_msgs/PoseStamped[] waypoint_poses

If a single-element list is provided as the goal, the resulting behavior will be that of the original move_base. The legacy interface is also available and consists of original topics move_base/goal, move_base/result, move_base/status, move_base/feedback, and move_base/cancel. The goal received on the legacy move_base interface will be type-converted and forwarded to the move_base_swp interface. The feedback and the result will be visible on both interface.

The goal received on the new move_base_swp interface will be executed natively and the feedback and result will be visible on the new interface only. Issuing the new goal on either of the two interfaces will preempt the goal in progress regardless of the interface the goal was issued on. Only one goal can be pursued at the time, regardless of the interface.

The simple-goal topic (/move_base_simple/goal) is also available for sending goals from RVIZ.

Navigation status topics

In addition to the actionlib interface, move_base_swp also provides several topics that can be used to monitor and visualize the progress of the goal. All topics that exist in the original move_base node exist in the move_base_swp and they are not repeated here.

The namespace used for all topics is move_base (not move_base_swp) because these topics are meant to be used together with original topics provided by the move_base.

The new topics are:

/move_base/current_waypoints (geometry_msgs/PoseArray)

This topic contains the list of waypoints that the robot is currently pursuing. As the robot visits each waypoint, it is removed from the list. The order of poses in the list is the order in which the waypoints will be visited.

/move_base/snapped_pose (geometry_msgs/PoseStamped)

This topic represents the robot pose (as derived from the transform tree) snapped to the global plan that the robot is pursuing. The snapping is done by finding the nearest point in 2D space that lies on the line composed out of dense waypoints generated by the global planner. This pose can be used to visualize the robot progress along the plan. Internally move_base_swp uses this pose to determine when to feed new segment of the global plan into the local planner (see Windowed global plan section for details).

/move_base/pursued_plan (nav_msgs/Path)

This topic is the subset of the global plan that has been fed to the local planner and that the robot is currently pursuing. As the robot makes progress along this plan, the waypoints are replenished from the global plan (published by the global planner).

Handbrake topic

/move_base/handbrake (move_base_swp/Handbrake)

Sending True to this topic will pull the handbrake stopping the robot along its plan. Sending False will release the handbrake. The handbrake must be renewed every second (or less) by re-sending True to the topic. This is required to prevent applications from halting the robot and disappearing. Note that this is not a safety brake (a safety brake would require an inverse logic that stops the robot if not renewed). This handbrake is meant to be used as part of the robot navigation strategy (e.g. stopping to yield to another robot).

Parameters

move_base_swp introduces several new parameters that control the behavior of the navigation state machine. All parameters are dynamic and can be changed at runtime with immediate effect or passed at node launch time. This section describes the available parameters. All parameters that are available in the original move_base node are available in move_base_swp with the same function. Their description is not repeated here.

~brake_slope (float, default: 0.5)

This parameter determines the deceleration rate (in m/s^2) that the robot will use when braking. Note that naturally slowing down when approaching the goal is not considered braking. Braking occurs when the goal is canceled, or preempted, or when the handbrake is pulled. Setting this parameter to a maximum value of 1000 (de-facto infinity) will effectively result in the original move_base behavior which is to stop immediately.

~brake_sample_rate (`float, default: 20)

Rate (in Hz) at which to generate velocity vectors when braking. This rate is used only when braking. During normal operation the rate specified by controller_frequency is used as the sampling rate.

~plan_buffer_size (int, default: 150)

The number of waypoints to keep loaded into the local planner when using Windowed global plan. Setting this parameter to zero will tun off this feature, effectively resulting in the original move_base behavior.

~plan_reload_threshold (int, default: 100)

The low-watermark threshold that determines when to load more waypoints into the local planner. Setting this parameter to a value that is greater than or equal to the value of ~plan_buffer_size will result in feeding waypoints continuously, which will result in least corner-cutting by the local planner, but may have performance implications. Setting this value to zero may result in choppy motion because the state machine will wait until the local planner exhausts all waypoints before supplying the new set.

~plan_min_step_len (float, default: 0.025)

This parameter ensures that the length of the plan buffer roughly results in the same covered distance for different map resolution. Namely, some global planner generate the waypoints whose resolution matches the map resolution. For example, 100 waypoints in the map whose resolution is 0.01 m/pixel would result in 10x shorter distance than in the map whose resolution is 0.1 m/pixel. Typically, such a fine resolution is not necessary because the local planner can successfully deal with waypoints that are much further apart. To deal with this problem ~plan_min_step_len parameter effectively downsamples the global plan such that the Euclidean distance between plan waypoints is greater or equal to the one specified by this parameter.

Usage

To use the move_base_swp node clone this repository into your catkin workspace and build it with catkin_make. Make sure that you have the ROS Navigation package installed and configure it to use the move_base normally. Make sure that works before cutting over to move_base_swp. Next, add additional parameters under move_base namespace in the same way you would set any other move_base parameter (either in your .launch file or in your parameter yaml file). If you don’t add new parameters, default values will be used which are fine to start with.

Next, modify your launch file for move_base and set the node type to move_base_swp. Start the move_base as you would normally start and your navigation system will run with move_base_swp. The relevant line in the .launch file should look like this:

<node pkg="move_base_swp" type="move_base_swp" name="move_base" output="screen">

Once the node starts up, you should see new topics and the new actionlib interface. Also, if you bring up rqt_reconfig tool, you should see the new parameters in the move_base section.

Changelog for package move_base

1.16.7 (2020-08-27)

• move_base: Add options for make_plan service (#981) Adds the following two parameters for the ~make_plan service: 1. make_plan_clear_costmap Whether or not to clear the global costmap on make_plan service call. 2. make_plan_add_unreachable_goal Whether or not to add the original goal to the path if it is unreachable in the make_plan service call.

• Contributors: nxdefiant

1.16.6 (2020-03-18)

1.16.5 (2020-03-15)

1.16.4 (2020-03-04)

• [Windows][melodic] Navigation (except for map_server and amcl) Windows build bring up (#851)
• Contributors: Sean Yen

1.16.3 (2019-11-15)

• Merge branch 'melodic-devel' into layer_clear_area-melodic
• Added publishZeroVelocity() before starting planner (#751) Edit for Issue #750
• Merge pull request #831 from ros-planning/feature/remove_slashes [melodic] Remove leading slashes from default frame_id parameters
• Remove leading slashes from default frame_id parameters
• Contributors: David V. Lu, Michael Ferguson, SUNIL SULANIA, Steven Macenski

1.16.2 (2018-07-31)

1.16.1 (2018-07-28)

1.16.0 (2018-07-25)

• Switch to TF2 #755
• Merge pull request #723 from moriarty/melodic-buildfarm-errors Melodic buildfarm errors
• Merge pull request #719 from ros-planning/lunar_711 adding mutex locks to costmap clearing service
• Contributors: Alexander Moriarty, Michael Ferguson, Vincent Rabaud, stevemacenski

1.15.2 (2018-03-22)

• Merge pull request #673 from ros-planning/email_update_lunar update maintainer email (lunar)
• Merge pull request #649 from aaronhoy/lunar_add_ahoy Add myself as a maintainer.
• Rebase PRs from Indigo/Kinetic (#637)
  • Respect planner_frequency intended behavior (#622)

  * Only do a getRobotPose when no start pose is given (#628) Omit the unnecessary call to getRobotPose when the start pose was already given, so that move_base can also generate a path in situations where getRobotPose would fail. This is actually to work around an issue of getRobotPose randomly failing.

  • Update gradient_path.cpp (#576)
  • Update gradient_path.cpp
  • Update navfn.cpp
  • update to use non deprecated pluginlib macro (#630)
  • update to use non deprecated pluginlib macro
  • multiline version as well
  • Print SDL error on IMG_Load failure in server_map (#631)
• Contributors: Aaron Hoy, David V. Lu!!, Michael Ferguson

1.15.1 (2017-08-14)

1.15.0 (2017-08-07)

• Add a max_planning_retries parameter to move_base [kinetic] (#539)
• Fix for #517: create a getRobotPose method on move_base instead of using that on the costmaps
• Fixed deadlock when changing global planner
• rebase fixups
• convert packages to format2
• Fix CMakeLists + package.xmls (#548)
• Added deps to amcl costmap_2d move_base (#512)
• Fix CMake warnings
• move_base: Add move_base_msgs to find_package.
• Contributors: Jorge Santos, Jorge Santos Simón, Maarten de Vries, Martin Günther, Mikael Arguedas, Vincent Rabaud, mryellow, ne0

1.14.0 (2016-05-20)

1.13.1 (2015-10-29)

• Removes installation of nonexistent directories
• use correct size for clearing window
• full name has been required for eons, this code just adds unneeded complexity
• remove ancient conversion scripts from v0.2 to v0.3
• proper locking during costmap update
• Contributors: Michael Ferguson, Thiago de Freitas Oliveira Araujo

1.13.0 (2015-03-17)

• Fixing various memory freeing operations
• Contributors: Alex Bencz

1.12.0 (2015-02-04)

• update maintainer email
• Contributors: Michael Ferguson

1.11.15 (2015-02-03)

• Disable global planner when resetting state.
• Mark move_base headers for installation
• Add ARCHIVE DESTINATION for move_base
• Break infinite loop when tolerance 0 is used
• remove partial usage of <tab> in the code
• Contributors: Gary Servin, Michael Ferguson, ohendriks, v4hn

1.11.14 (2014-12-05)

• use timer rather than rate for immediate wakeup
• adding lock to planner makePlan fail case
• Contributors: Michael Ferguson, phil0stine

1.11.13 (2014-10-02)

1.11.12 (2014-10-01)

1.11.11 (2014-07-23)

• removes trailing spaces and empty lines
• Contributors: Enrique Fernández Perdomo

1.11.10 (2014-06-25)

• Remove unnecessary colons
• move_base planService now searches out from desired goal
• Contributors: David Lu!!, Kaijen Hsiao

1.11.9 (2014-06-10)

• uses ::hypot(x, y) instead of sqrt(x*x, y*y)
• Contributors: Enrique Fernández Perdomo

1.11.8 (2014-05-21)

1.11.7 (2014-05-21)

• update build to find eigen using cmake_modules
• Fix classloader warnings on exit of move_base
• Contributors: Michael Ferguson

1.11.4 (2013-09-27)

• Package URL Updates
• Reintroduce ClearCostmaps Service
• Add dependencies to recovery behaviors.