Deep Implicit Imitation Reinforcement Learning in Heterogeneous Action Settings

Authors: Iason Chrysomallis, Georgios Chalkiadakis, Ioannis Papamichail, Markos Papageorgiou

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results confirm that the benefits associated with deep implicit imitation, namely accelerated training and improved performance over the mentor agent, are achieved even in contexts with non-homogeneous action settings. Our experiments are conducted both in a maze environment similar to those used for the study of tabular implicit imitation RL (Price and Boutilier 1999, 2001, 2003), and in a challenging forceactuated navigation environment (Fu et al. 2020).
Researcher Affiliation Academia Iason Chrysomallis, Georgios Chalkiadakis, Ioannis Papamichail, Markos Papageorgiou Technical University of Crete, Chania, Greece EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Deep k-n step repair
Open Source Code No The paper mentions a technical appendix for additional experiments but does not explicitly state that the source code for the methodology described in the paper is available or provide a link to a code repository.
Open Datasets Yes Specifically, we utilize the 2D Maze environment from Open AI Gym (Brockman et al. 2016), introducing a maze of size 30 30 where the state space comprises the agent s current position. Specifically, we employ D4RL s Point Maze (Fu et al. 2020), an environment based on the Mu Jo Co simulation physics engine (Todorov, Erez, and Tassa 2012)
Dataset Splits No The paper describes how episodes are defined and when an environment is considered solved, but it does not provide specific train/test/validation dataset splits for reproducibility.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions using frameworks like DQN, OpenAI Gym, and MuJoCo, but does not provide specific version numbers for these or any other software libraries or programming languages used.
Experiment Setup No The paper mentions hyperparameters like 'einfeas' and search depths 'k' and 'n', but describes their selection as 'domain-specific' and suggests 'low values' without specifying the exact numerical values used in their experiments. It does not provide details on learning rates, batch sizes, optimizers, or training epochs.