EMOTE: An Explainable Architecture for Modelling the Other through Empathy

Authors: Manisha Senadeera, Thommen Karimpanal George, Stephan Jacobs, Sunil Gupta, Santu Rana

IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experiment on minigrid environments showing EMOTE: (a) produces more consistent reward estimates relative to other IRL baselines (b) is robust in scenarios with composite reward and action-value functions (c) produces human-interpretable states, helping to explain how the agent views other agents. We conducted two sets of experiments to assess EMOTE s performance
Researcher Affiliation Academia Manisha Senadeera1 , Thommen Karimpanal George 1,2 , Stephan Jacobs3 , Sunil Gupta1 and Santu Rana1 1Applied Artificial Intelligence Institute, Deakin University 2School of Information Technology, Deakin University 3University of Queensland
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Supplementary materials available1. 1 https://github.com/manishasena/EMOTE
Open Datasets No The paper mentions 'minigrid environments' and 'Environments were designed using Marl Grid [Ndousse, 2020] based on Mini Grid [Chevalier-Boisvert et al., 2018].' These references describe the simulation environment/framework used for training agents, not a specific publicly available dataset with concrete access information for data used in the experiments.
Dataset Splits No The paper does not provide specific percentages or counts for training, validation, or test dataset splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions 'DQN [Mnih et al., 2015]' for policy learning and 'Marl Grid [Ndousse, 2020] based on Mini Grid [Chevalier-Boisvert et al., 2018]' for environment design, but it does not specify version numbers for any of these software components or other libraries used for implementation.
Experiment Setup Yes The hyperparameter δ [0, 1] balances the importance of reconstructing the empathetic state se to be similar to the original state si (interpretability) and the accuracy of the learning agent s predicted actions in se. This was applied to 3 Agent game (ψ = 5e 4).