Refactoring Python/C++ redundancy

[stulp]

Rationale for removing bbo C++ implementation

There is obvious code redunancy due to the parallel Python and C++ implementations. This is because there are two different purposes.

• C++ for real-time implementation, and completeness (implements many more features)
• Python better for scripting and didactic purposes.

The modules bbo and dmp_bbo need not be real-time, and therefore the C++ implementation is not really necessary.

Planned refactoring steps

1. Port functionality from src/bbo to python/bbo (and dmp_bbo also)
2. Add docstrings in Python from C++ code
3. Apply consistent formatting in Python with black
4. Remove src/dmp_bbo, src/bbo

Further steps along the way:

• Rename dmpbbo_io => eigen_tools
• Get rid of boost::serialization, which is very annoying, and replace with a YAML/JSON library?

[stulp]

Further idea: anything that is not real-time need not be in C++ (e.g. training)

Use case:

• train dmp (in Python) and write to file (using Protobuf)
• run dmp real-time on the robot (in C++) by reading it first with ProtoBuf

All train() functions could then be removed from the C++ version (i.e. also no more meta-paramers necessary in C++)

In C++, only the UnifiedModel would be necessary (or perhaps two, i.e. weighted basis functions or weighted line segments)

[stulp]

The redundancy has been resolved in v2 of dmpbbo: https://github.com/stulp/dmpbbo/tree/v2