Subequivariant Graph Reinforcement Learning in 3D Environments

Authors: Runfa Chen, Jiaqi Han, Fuchun Sun, Wenbing Huang

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we evaluate the proposed method on the proposed benchmarks, where our method consistently and significantly outperforms existing approaches on single-task, multi-task, and zero-shot generalization scenarios. Extensive ablations are also conducted to verify our design.
Researcher Affiliation Collaboration 1Dept. of Comp. Sci. & Tech., Institute for AI, BNRist Center, Tsinghua University 2THU-Bosch JCML Center 3Gaoling School of Artificial Intelligence, Renmin University of China 4Beijing Key Laboratory of Big Data Management and Analysis Methods.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code and videos are available on our project page: https://alpc91.github.io/SGRL/. and Our codes are available on https://github.com/alpc91/SGRL.
Open Datasets Yes The environments in our 3D-SGRL are modified from the default 2D-planar setups in Mu Jo Co (Todorov et al., 2012). Specifically, we extend agents in environments including Hopper, Walker, Humanoid and Cheetah (Huang et al., 2020) into 3D counterparts.
Dataset Splits No We keep 20% of the variants as the zero-shot testing set and use the rest for training.
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 "Py Torch" and "Mu Jo Co" as software used, but does not provide specific version numbers for these or any other key software dependencies.
Experiment Setup Yes Table 6 provides the hyperparameters needed to replicate our experiments. Table 6 lists hyperparameters such as "Learning rate 0.0001", "Mini-batch size 100", and "Total attention layers 3".