This package integrates trajectory smoothing methods in the ROS framework.
The input is a list of 3D points which correspond to a movement with initial and final velocity equal to zero (reaching motion). The functionality of the package is to use the selected smoothing function (Bezier Curves or Smoothing Splines) in order to extrat a smooth trajectory from the input 3D points. In addition, an empirical statistical method is used based on the profiles of a reaching motion in order to excude the points that are concentrated in the beginning and ending of the input movement which do not offer any usefull information about the actual trajectory.
In order to implement the smoothing function this repo is using soucre based on the following repos:
- Bezier Curves smoothing: https://github.com/Hrisi/Python---Spline-curves
- Smoothing splines: a) Initial off-ROS implementaton (in Python3): https://github.com/espdev/csaps b) Sklearn-based implementation (+ Polunomial smoothing): https://github.com/madrury/basis-expansions
The above above smoothing fuctions are offered as a ros-service by the smooth_server.py script (server). Default option is the Bezier Curve based smoothing.
This package was tested using as input 3D points that represent reaching motions of the human wrist inside the workspace of a UR3 robotic arm which where tracked by a RGB-D camera using Openpose as part of the following pipeline (currently not integrated):