Multi Time Scale World Models

Authors: Vaisakh Shaj Kumar, SALEH GHOLAM ZADEH, Ozan Demir, Luiz Douat, Gerhard Neumann

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate our approach to a diverse set of simulated and real-world dynamical systems for long-horizon prediction tasks. Our experiments are designed to answer the following questions. (a) Can MTS3 make accurate long-term deterministic predictions (mean estimates)? (b) Can MTS3 make accurate long-term probabilistic predictions (variance estimates)? (c) How important are the modelling assumptions and training scheme?
Researcher Affiliation Collaboration 1Karlsruhe Institute Of Technology (KIT), Germany 2SAP SE, Germany 3Robert Bosch Gmb H, Germany
Pseudocode No No, the paper describes its methods through text and mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code is available at this repository: https://github.com/ALRhub/MTS3.
Open Datasets Yes (a) D4RL Datasets We use a set of 3 different environments/agents from D4RL dataset [6], which includes the Half Cheetah, Medium Maze and Franka Kitchen environment.
Dataset Splits No No, the paper discusses training and evaluation but does not provide specific percentages or sample counts for training, validation, and test dataset splits.
Hardware Specification No No, the paper does not provide specific hardware details such as GPU models, CPU specifications, or memory amounts used for running experiments.
Software Dependencies No No, the paper does not list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch x.x, or specific library versions).
Experiment Setup Yes Finally, we perform ablation for different values of H. t, which controls the time scale of the task dynamics. The results reported are for the hydraulics dataset. The higher the value of H, the slower the timescale of the task dynamics relative to the state dynamics. As seen in Figure 4a, smaller values of H (2,3,5 and 10) give significantly worse performance. Very large values of H (like 75) also result in degradation of performance.