资源算法xQP-tasks-constraints

xQP-tasks-constraints

2020-02-26 | |  41 |   0 |   0

xQP-tasks-constraints

The tasks and constraints that are the building blocks of a generic QP solver

This repository contains the code for individual tasks and constraints that constitute a QP stack (e.g., iHQP). Each task or constraint will compile into a single shared object (.so) that are loaded by the iHQP solver. The rational for this decision is outlined later in this readme.

Dependencies

For the moment:

  1. Eigen3 (for linear algebra) http://eigen.tuxfamily.org

Motivation

Goals

The implementation follows few goals:

  1. Keep it light-weighted, easy to use, easy to deploy

  2. Reduce the dependencies to bare minimum

  3. Make it compliant and easy to adjust if the end user wants to use it

    • as a binary

    • in a component based system (e.g., OROCOS, Microblx, etc.)

    • by conforming to certain standards

  4. In short, don't annoy the users!

Reinventing the wheel

There are already many many qp based packages for robot control by different research groups with exceptionally high quality implementations. However, most of these works are rather tailored to specific solver and/or theoretical approach. Special to this implementation is OpenSoT by (mainly) Enrico Mingo HoffmanAlessio Rocchi and Arturo Laurenzi from the Italian Institute of Technology.

This work is heavily inspired by OpenSoT and its very nice structure. However, for few reasons it tries to implement OpenSoT's idea in different manner. Namely,

  1. OpenSoT is highly coupled with several other packages that are not necessarily critical for the end users. This makes its usage on stand-alone applications rather cumbersome.

  2. OpenSoT compiles its very rich collection of tasks and constraints into one binary. This implementation on the other hand goes for a different idea (more on this below)

  3. While OpenSoT, its sibling projects (e.g., CartesI/O) and all its dependencies are available for free as a debian package, for the time being it's not open source. Consequently, it is not so trivial to reshape and refactor it to make it compatible to specific requirements.

  4. OpenSoT's configuration is rather exhausting and depends on an unhealthy number of yaml files. This is not OpenSoT's fault per se, rather due to its high number of dependencies and to number 1 in this list.

On this implementation

Current work borrows the idea of task/constraint library from the OpenSoT but compiles each of them into individual .so files. A particular QP based solver/implementation then loads these tasks/constraints on demand. Hence, the users are not forced to adopt anything forced by us.

Such approach that mimics the component paradigm, grants some conveniences. i) The code generation complexity for the iHQP solver (or any other approach for that matter) will be reduced considerably; ii) Stand-alone use of this library for the community is also facilitated as we do not enforce any limitation on the end-users; iii) Composability and reusability are alleviated; and iv) It facilitate the conformity to particular standards, should the library is used for specific projects.

On RobMoSys

This project is being developed in the context of an ITP in RobMoSys project. Consequently it should adhere to RobMoSys standards and it should conform to it. The code generation aspect shall be taken care of in the iHQP implementation. As for the conformity, once it is decided what it exactly means, a separate branch will be taking care of it. It is my goal to keep the master branch as a generic, stand-alone, ready to be used set of tasks and constraints.

TODO

  1. Give it a name!

  2. In the spirit of code first, think later (!) I am adopting a structure that seems reasonable. However, the structure of the repository should be rethinked if necessary.

  3. Consider suitable namespaces. It is my goal to limit the use of namespaces to the absolute necessary only. I do not wish to create eye sores for the users just so that I put my library name in front of their eyes.

  4. Consistency check for calling them task or constraint. As it was said by Kanoun et al. "To call these equations constraints or tasks is just a question of context.", however, it is necessary to be consistent regardless.

  5. TBA


上一篇:Core-Maths-QPS

下一篇:deeppose_tf

用户评价
全部评价

热门资源

  • seetafaceJNI

    项目介绍 基于中科院seetaface2进行封装的JAVA...

  • spark-corenlp

    This package wraps Stanford CoreNLP annotators ...

  • Keras-ResNeXt

    Keras ResNeXt Implementation of ResNeXt models...

  • capsnet-with-caps...

    CapsNet with capsule-wise convolution Project ...

  • inferno-boilerplate

    This is a very basic boilerplate example for pe...