Reinforcement learning Q Learning Grid-World 8 directions. Change number of grids, position of trap, goal in the gridworld.py
QTable.ipynb is a colab notebook for implementing Q-learning in Gridworld with 8 action space with visualisation. Qtable_gridworld.ipynb is the whole notebook which is not implemented in colab.
One of the reasons people love playing games is the sense of agency. They have control over their choices and the outcomes are instantaneous, causation obvious. For an average human who makes thousands of choices, causations are poorly understood as the processes can be abstract. Simple everyday nuances can compound over time. Even if people know the cause and effect, do they have the power to change it?
For a person who is disabled their choices are fundamentally restricted by sensorimotor impairment, causing low feeling of agency, in everyday activities. Especially in stroke where they were once independent. Competence, Autonomy or sense of agency/control are fundamental psychological needs. With increase of perceived competence, intrinsic motivation increases, yet competence alone is not enough as the perception that they won the game by their effort is vital, which is the feeling of agency. As feeling of agency decreases, intrinsic motivation decreases
The main rational is to give assistance without compromising the feeling of agency/control. In this simulation a simple gridworld game was agent trained using Q-learning algorithm, the assistance is compromised between the feeling of agency/control and competence(score). Feeling of agency in this case depends when the assistance is made. If the optimal angle is entirely opposite to subject's angle and yet the assistance is given the feeling of agency will be low, competence (score) is high. We would like to increase feeling of agency/control and competence at the same time, which are two opposite parameters when assistance comes into play. A tradeoff between the feeling of agency and competence is vital for keeping subject engaed in the game.
Here three different weights simulations results are shown,
Assist As Needed = Competence 1.0 + Autonomy .75
Assist As Needed = Competence 1.0 + Autonomy 1.0
Assist As Needed = Competence 0.75 + Autonomy 1.0
P.S: The Qtable_gridworld.ipynb is in a raw unrefined stage, will upload a clean code in the near-future. Simulated weighted assistance vs game score