Skills Regularized Task Decomposition for Multi-task Offline Reinforcement Learning

Authors: Minjong Yoo, SangWoo Cho, Honguk Woo

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through experiments, we show that our multitask offline RL approach is robust to the mixed configurations of different-quality datasets and it outperforms other state-of-the-art algorithms for several robotic manipulation tasks and drone navigation tasks.
Researcher Affiliation Academia Minjong Yoo, Sangwoo Cho, Honguk Woo Department of Computer Science and Engineering Sungkyunkwan University {mjyoo2, jsw7460, hwoo}@skku.edu
Pseudocode Yes Algorithm 1 Skill regularized task decomposition
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]
Open Datasets Yes For evaluation, we use several robotic manipulation tasks and drone navigation tasks using the Meta-world environment [26] and the Airsim drone simulator [27].
Dataset Splits No The paper describes the types of datasets used (MR, RP, ME) and the number of episodic trajectories in each (150, 100, 50), and mentions using 'mixed configurations of different datasets' for evaluation. However, it does not explicitly provide information on train, validation, or test dataset splits in the main text, stating that 'The detailed settings including hyperparameters and environment conditions can be found in Appendix.'
Hardware Specification No The ethics statement indicates that the paper includes 'the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)', but these specific hardware details are not provided within the main body of the paper.
Software Dependencies No The paper does not explicitly mention specific software dependencies with version numbers, such as Python versions or library versions.
Experiment Setup Yes The detailed settings including hyperparameters and environment conditions can be found in Appendix.