Only released in EOL distros:
A package for learning task relevant features. Contain behaviors for operating flip style light switches, rocker style light switches, and drawers.
- Author: Hai Nguyen (email@example.com) Advisor: Prof. Charlie Kemp, Lab: Healthcare Robotics Lab at Georgia Tech
- License: BSD
- Source: git https://code.google.com/p/gt-ros-pkg.hrl/ (branch: master)
This package uses visual feature learning to create detectors for operating the behaviors included (drawer opening, flip style light switches, and rocker style light switches). Before using the included behaviors on real world mechanisms, you'll first need to perform an initialization and a data collection step to create a visual classifier for your specific mechanism. After these two steps, you'll be able to run the included behaviors with your new learned visual detector using the execute launch script (execute.launch).
Specifically, the steps that you'll need to perform are:
- launch navigation and manipulation nodes required (trf_learn.launch)
roslaunch trf_learn trf_learn.launch
- Localize the robot with rviz.
- Position your robot in front of the mechanism to operate.
- Initialize the learning procedure.
roslaunch trf_learn init.launch
- * after this completes the collected data will be saved in ~/.ros/locations_narrow_v11.pkl
- Launch the learning procedure.
roslaunch trf_learn practice.launch
Finally to execute your learned behavior use:
roslaunch trf_learn execute.launch
When initialization finishes, you'll progress to the second stage and set the robot to gather additional data autonomously for its classifier (using practice.launch). Due to the experimental nature of the code, currently this data collection process takes a few hours but once it's done use the task execution script (execute.launch) to activate the behavior with its trained detector.