I resume my work on thesis in the end of august by implementing advanced logging into my C++ projects.
Then, I wrote a python script so I have infrastructure to collect data from experiments and store them in my “experiments library“.
I made several experiments with various parameters of CACLA algorithm and Neural networks architecture (for Actor and Critic) to examine different behaviour.

I had also finished class ICubGrasping which is used to manipulate with iCub’s hands to perform operations which will be executed while grasping.

Now, I have to think out first basic architecture for simple object grasping task which I could implement and test on iCub.
Then, after I get gladsome results I can go further to develop more advanced and flexible architecture.

After I finished all my exams and took a rest for a couple of weeks I returned to my master thesis.
So what did I do in this period:

• added separate neural networks that acts as forward and inverse model, which both learns from data gathered during CACLA learning (to reach object).
• created a C++ class with a lot of helpful commands for iCub (methods for environment construction, body manipulation, etc.)
• created class which specializes on grasping (hand manipulation) with parametrized methods to hand flexion, extension and such things.

I plan to connect this grasping class to some learning algorithm (like CACLA) and experiment with it.

I also read more carefully an article about Infant grasp learning and found out lots of interesting biological information (maybe I can later compare my experiment with iCub to how human infants learn to grasp).