Robust Predictable Control

Authors: Ben Eysenbach, Russ R. Salakhutdinov, Sergey Levine

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments have two aims. First, we will demonstrate that RPC achieves better compression than alternative approaches, obtaining a higher reward for the same number of bits. Second, we will study the empirical properties of compressed policies learned by our method, such as their robustness and ability to learn representations suitable for hierarchical RL.
Researcher Affiliation Collaboration Benjamin Eysenbach1 2 Ruslan Salakhutdinov1 Sergey Levine2 3 1Carnegie Mellon University, 2Google Brain, 3UC Berkeley
Pseudocode No The paper describes the algorithm in text and refers to “Appendix B and the open-sourced code for details” but does not include pseudocode in the main paper.
Open Source Code Yes Project site with videos and code: https://ben-eysenbach.github.io/rpc
Open Datasets Yes We evaluate all methods on four tasks from Open AI-Gym [5] and two image-based tasks from dm-control [40].
Dataset Splits No The paper does not explicitly provide specific percentages or counts for training, validation, and test dataset splits.
Hardware Specification Yes The image-based experiments used half of a NVIDIA p100 or v100 GPU. The state-based experiments did not use a GPU.
Software Dependencies No The paper mentions software components like Open AI-Gym, dm-control, TF-Agents, and SAC, but it does not provide specific version numbers for these or other software dependencies.
Experiment Setup No We refer the reader to Appendix B and the open-sourced code for details. The main text does not contain specific hyperparameter values or detailed training configurations.