After rebasing on the latest version of BT.CPP, create a feature branch with both the contents of

• glpuga/BehaviorTree.CPP#5
• glpuga/BehaviorTree.CPP#6

To build on top of that, since the mocking feature might come handy.

Tasks

• Create branch with commits from both and test it.
• Set the submodule in this repo to track that branch.

Add a package with tools to streamline working with BTs:

• #67
• #66
• #68
• #72
• #76

• Pull changes from upstream
• Rebase feature branches on the current devel head.

A few of the exceptions there are probably no needed or indicate actual issues in the code.

Create BehaviorTreeManager class that fully wraps a BT,

• It gets created from the BT::BehaviorTree object.
• It must provide a simple run() method to run-to-completion the tree (until the root returns either Success or Failure).
• The class must take care of ticking and threading internally.

Linked to: #65

Add my fork of BehaviorTree.cpp to the external projects as a submodules.

https://github.com/glpuga/BehaviorTree.CPP/

Follow up on #55 . See there for more information.

Create an inheritable BehavorTreeFactoryInterface with a partial BehavorTreeFactory.

• Create factory
• Create factory test

After #36 check if tijcore still needs a direct dependency on Eigen. It shouldn't.

The current version of the mocking feature for BT inherits the mocking node from SyncActionNode, which prevents the mock from returning RUNNING because SyncActionNode tests for that value and throws if an attempt is made to return it.

This can be improved by inheriting directly from BaseActionNode instead.

Running roslaunch nist_gear sample_environment takes several minutes to actually launch the simulation, but eventually succeeds.

This impairs our ability to iterate solutions and code quickly.

At least part of the time seems to be due to gazebo downloading models to .gazebo/models . This however, does not seem to substantially reduce the time it takes to launch the simulation a second time on the same container.

This ticket is about looking for ways to reduce the load time, preferably to tenths of seconds.

There's a test branch testing Industrial CI to this repo.

• Research a bit more industrial CI
• Setup code to a workable standard for use with ICI.
• Setup or discard GA as CI
• Measure consumption, to determine what the best settings for running CI is (2000 minutes monthly limit)

Requires:

• #25 to prepare the codebase, which was never properly prepared for this during the competition.

Currently Inductrial CI in the self-hosted runner is configured to use whatever number of cores is available, but that causes peaks of memory use when building the repository.

This ticket is about limiting the number of cores to about half. The build time will not be too affected, since most of the process is executed sequentially anyway.

Following instructions from last year's competition here:

https://github.com/usnistgov/ARIAC/blob/ariac2021/wiki/tutorials/moveit_interface.md#using-rviz

A launch file could be created to use rviz to control the robot arms.

This ticket is about:

• Create package for utils
• Create a launch file for launching rviz.
• Document how to launch it in README.md.

Ccache is great, but the first fuill build still takes the full time each time that the docker is started.

This ticket will research mounting the cache on a volume so that it persists across docker launches.

Install catkin-tools to allow building using catkin build as an alternative to catkin_make and catkin_make_isolated.

Remove the logger to a separate package. It should not be part of tij_core.

Documentation

• Ccache documetantion: https://ccache.dev/manual/4.5.1.html
• Step by step to use with ROS: https://jquintas.medium.com/compile-faster-in-ros-ccache-9d13ff0bff44

It was also necessary to set CC and CXX so that catkin_make could find ccache.

Configurations needed

• Install ccache in ubuntu docker image.
• Add /usr/lib/ccache to the start of the PATH so that calls to the compiler go through ccache.
• Set CC="gcc" and CXX="g++" so that catkin_make go through ccache instead of using absolute paths to the compilers.

As great as it is, Industrial CI takes a long time to run, most of which it spends installing the dependencies.

For PRs it would be great to have something faster, like just building the devel docker image and using it to build and run the tests. This would be much faster with my setups, because It will be faster, my self-hosted runner will cache docker image layers and make both building the image and running the tests very fast in the average case.

Pro:

• That's exactly the devel environment I'm using. ICI is bulding one from the dependencies, but that does not mean my docker image is correct since it does not install the dependencies.
• Much faster.

Con

• Dependencies should be sufficient to install the system, and testing from the docker does not ensure that that's the case because it redundantly lists the required ROS packages instead of using rosdep to generate the list.

Proposal:

• Build and run from the docker for PRs. Build and run from industrial CI for pushes to devel/master.
• Create separate runners, one for fast track (PRs) and another one for a slow track (Industrial CI). this will avoid slow builds to cause queueing for PRs.
• Create a ticket to update docker image to install all dependencies using a list generated with rosdep.

Configure repository for a proper master/devel/feature branch workflow with rebase merge by default.

Most of the codein tijbt was moved to a separate repository for reusability within another project.

See https://github.com/glpuga/BehaviorTreeExtras for the new location of the code.

This ticket is about:

• Remove the parts of tijbt that were relocated to the new dependency. This is mostly all, except what's related to the logger.
• Add external repository as a subrepo in "external".
• Refactor code to use the external project for the logger tests.

Create a docker image to be able to run the simulation consistently using the VORC workspace dockerfiles as starters.

Linked to: #65

devutils is the first package that was created, following a file struture that was not completed later.

Remove or refactor into other folders.

The current subtree mocking approach works, but leaves a lot of boilerplate on the hands of the user to actually set the mocks, since the user needs to demultiplex the calls the SubtreeMock node in the node builder.

See example in tijbt/tests, in the factory class test, and tests in the behaviortree.cpp fork.

The current approach of creating a builder lambda that multiplexes at instantiation time based on the name of the subtree needs to be revised and/or simplified by providing a ready to user solution for this.

Ever since I used structured bindings in the ManangedLocus interface to return model info, I've regretted it. Most often than not the key part of the pair is often not used, and causes "variable declared but not used errors".

Update the inter

• Update the interface to return the part id and the broken flag separately.
• Update tests.

The current folder structure in tijmath and tijcore is totally meaningless.

This issue will address tijmath and flatten it.

• This test has been failing since the last push from ARIAC 2021, when it broke after adjusting the algoritm for the resource manager.
• The resource manager is likely to be completely rewritten, so this ticket is about either deleting this test, or fixing it so that it passes.
• For the time being it's disabled to get CI up and running.

The test measures execution time for the halt() command to take effect, but it seems to be taking longer than expected from time to time. Investigate and fix.

• Great data on using PDDL for surveing, including a comparison of the main available solvers: https://repository.tudelft.nl/islandora/object/uuid:5bc6b1f4-1acf-4b40-8154-828427f9a4e9/datastream/OBJ/download

Decuople creating a BT logger from creating the logger, so that loggers can be attached to a tree at some point after creation, making possible injecting them.

• Create an attachToTree() method in AbstractLogger.
• Refactor BT code to attach to tree using the new method.
• Update BT tests.
• Update ariac2021 code to use the new addition.

#8 changed the commit the ARIAC submodule is pointing to, to address an error in a dependency in the ariac package in that repo.

This issue should remove that fix and restore the submodule to point to the master branch in the fork, which should in time be synced to the master branch of the official repo.

Tasks:

• Sync https://github.com/glpuga/ARIAC with the official repo, https://github.com/usnistgov/ARIAC
• Restore the branch pointed to by the submodule to the current master.

Idea: An environment variable might work to set for both tests and executable nodes.

Add a Readme.md file with instructions to:

• download the repo including submodules
• build the container
• launch the container
• join the container
• build the workspace
• launch the sample environment

• Dusting the repository
• Setting dependencies
• Adding roslint

The ARIAC upstream repository seems to have wrong dependency declared, which gets caught by rosdep.

This issue will fix it until upstream gets fixed.

diff --git a/ariac/package.xml b/ariac/package.xml
index b77e085..0fc99ba 100644
--- a/ariac/package.xml
+++ b/ariac/package.xml
@@ -11,7 +11,7 @@
   <exec_depend>nist_gear</exec_depend>
   <exec_depend>test_ariac</exec_depend>
   <exec_depend>ariac_example</exec_depend>
-  <exec_depend>gantry_moveit_config</exec_depend>
+  <exec_depend>gantry2021_moveit_config</exec_depend>
   <buildtool_depend>catkin</buildtool_depend>
 
   <export>

• Update submodule
• Add any additional dependencies in docker
• Check that demo scenarios launch