Porting code from v1.x

Some of the differences w.r.t. version 1.x are summarized hereafter.

General design

  • Almost all functionalities of CartesIO are now provided task-wise, rather than from a global CartesianInterface object as before. The old API has been preserved for backward compatibility. A task object is obtained by name with the getTask() method, and is then dynamically casted with std::dynamic_pointer_cast.

  • The most part of the implementation has been moved to task-specific classes implementing

    • reference management (e.g. time-out, filtering, trajectory generation, …)

    • ROS API

    • translation into OpenSoT tasks/constraints

    • …

  • A Subtask task type has been introduced to distribute parts of a task over different priorities

  • Richer API

  • The ControlMode property which could task values among Position, Velocity, Disabled is now splitted into

    • an ActivationState which is defined for all tasks, and can be either Enabled or Disabled

    • a ControlMode which is defined for Cartesian tasks only, and can be either Position or Velocity

ROS

  • task_name/state topic has been renamed to task_name/current_reference

  • world_odom TF frame has been removed, and the standard world is used instead

C++

  • The ROS Client has been renamed, from RosImpl to RosClient. The same applies to the header file name.

  • A new Context object has been introduced to gather together a struct of Parameters (among which the control rate), and the ModelInterface

  • Almost all constructors which would take a ModelInterface an input, now take a Context shared pointer

  • Implementations of CartesianInterfaceImpl are now created with the static factory ::MakeInstance, rather than manually looking for the shared object and invoking its extern "C" factory method by name

Python

TBD