Transfer Learning with Probabilistic Mapping Selection
Anestis Fachantidis, Ioannis Partalas, Matthew E. Taylor, Ioannis Vlahavas
Videos
The following videos show iCub learning to hit the ball in the iCub simulator, Both versions of the task use the novel YARP wrapper for RL-Glue.
Source Task
iCub learns to hit the ball by only moving two of its torso joints. The terminal window shows the number of the learning episode, the reward obtained and the number of joint moves that were needed to hit the ball.
Target Task
iCub learns to hit the ball by moving four of its shoulder joints. The terminal window shows the number of the learning episode, the reward obtained and the number of joint moves that were needed to hit the ball.
Source Code
Here you can find the first version of our novel YARP wrapper for RL-Glue. Together you will find the source code for the rl-glue environment of the target task, which needs the YARP wrapper to compile.
Please
contact Anestis Fachantidis for bug reports, comments, suggestions or request
for help with the source code. A new and cleaner version of the source code will be soon uploaded to rl-library