Adapted my old/archived my_matlab_rosbag code to work with the ROS Toolbox for MATLAB (v2019 and up).

1. Record Demonstrations as ROSbags: Record demonstrations as time-series of geometry_msgs::PoseStamped (for position and orientation, WrenchStamped can be used for FT messages) messages in rosbags. A repo that provides useful code (for different robots) and launch files in ROS for easy data collection can be found in the easy-kinesthetic-recording package.
2. Extract Trajectories from ROSbags to MATLAB: Extract desired messages and create trajectories of matlab structures (containing all pose information) stored in .mat. See examples below.
3. Segment Trajectories: In this repo you will find the scripts to define segmentation points when gripper state has opened/closed -- this is only useful for pick-and-place tasks! For continuous tasks that are not defined by gripper changes a seperate segmentation algorithm must be used, such as the one being developed in dsltl which is used for some of the tasks below.

The script xsens_mitsubishi_process_rosbags.m loads the rosbags that were recorded by tracking the right hand of a human operator with the Xsens MVN Motion Capture system and workspace object locations via a dual-kinect setup using the AprilTags detection algorithm as designed for the Mitsubishi Cobot Assista MIT Project.

Note: This script only extracts the trajectories from the saved rosbags and converts them to the robot's reference frame.

Trajectory Segmentation: In order to learn individual goal-oriented motion policies (such as attractor-based Dynamical Systems) from this data one must segment the trajectories. For this task a state-change segmentation algorithm is used to cluster trajectories corresponding to different known action proposition (AP) regions. This approach is under development/in preparation for submission by Felix Yanwei Wang and Nadia Figueroa, see dsltl.

This task involves scooping and mixing ingredients from bowls. The script franka_cooking_process_rosbags.m loads the rosbags recorded by tracking the end-effector for the franka emika panda during kinesthetic demonstrations as shown below:

This kinesthetic teaching example is documented in the franka_interactive_controllers package or the easy-kinesthetic-recording package on the latest-franka branch.

Note: The franka_cooking_process_rosbags.m script only extracts the trajectories as shown above.

Trajectory Segmentation: In order to learn individual goal-oriented motion policies (such as attractor-based Dynamical Systems) from this data one must segment the trajectories into clusters as shown below:

The same segmentation algorithm for the industrial task is used here. See dsltl or contact Felix Yanwei Wang and/or Nadia Figueroa.

Involves grasping plates/cutlery from dish rack and placing it on a table. The script franka_tablesetting_process_rosbags.mloads the rosbags recorded by tracking the end-effector for the franka emika panda during kinesthetic demonstrations as shown in the kinesthetic teaching example with the franka_interactive_controllers package or the easy-kinesthetic-recording package on the latest-franka branch.

Gripper State-Based Trajectory Segmentation Since these continuous demonstrations involving a series of pick-and-place tasks we can automatically segment the trajectories into goal-oriented clusters based on the state of the of gripper and the object locations where the gripper grasped/released an object.

To fill:...

Nadia Figueroa (nadiafig AT seas dot upenn dot edu)