CHI2021-Gaze-Based-Selection
We built the gaze-based selection task as an OpenAI gym like environment (see folder 'envs') and used the Stable Baselines3's implementation of the deep RL algorithm, Proximal Policy Optimization to train the model (https://github.com/DLR-RM/stable-baselines3). Hyperparameters were as follows: Horizon=$500$, Clipping:=$0.15$, Gamma=$0.99$. Other hyperparameters for the model are defaults as in Stable Baselines3 implementation. The hyperparameters for the human bounds used in the model is provided in Table 1.