This library provides a PDDL-to-C++ conversion tool that allows you to decrease runtime by "compiling" PDDL (Planning Domain Definition Language) text into a C++ interface. By converting PDDL into C++, you can benefit from the performance optimizations and compile-time checks provided by the C++ programming language.
PDDL Executor: The library includes functionality to generate C++ behavior trees from PDDL with sensing actions.
The pipeline is broken into several phase. First, a planning problem and domain must be constructed with PDDL.
Then the PDDL is parsed with provided the Python package and converted into a C++ with codegen.
The generated code implements an expand method to generate successor state from an initial state.
It also provides a initialize_problem function to help setup the problem instance. A planner is provided,
but it would be easy to reuse the basic framework with a custom planner.
Once a plan is found at runtime, it is converted into a behavior tree encoded and then executed. Notable, code generation is used again to create a C++ interface that is used by the behavior tree. Essentially, the interface contains function for all actions in the PDDL domain, which need to be implemented.
Run the following command from the project root to generate a sample plan. Note the first run is slower because it compiles the PDDL.
ros2 run plan_solver_py plan_solver -o $(pwd)/pddl_executor_example/pddl/domain_blocks.pddl -f $(pwd)/pddl_executor_example/pddl/problem_blocks.pddl
Then run the following command to visualize the plan graph.
ros2 run plan_solver_py plan_graph
You should see the following plan generated from the block world domain.
A complete example with an explanation can be found here