Object-Oriented Dynamics Learning through Multi-Level Abstraction
Authors: Guangxiang Zhu, Jianhao Wang, Zhizhou Ren, Zichuan Lin, Chongjie Zhang6989-6998
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our results show that MAOP significantly outperforms previous methods in terms of sample efficiency and generalization over novel environments for learning environment models. We also demonstrate that learned dynamics models enable efficient planning in unseen environments, comparable to true environment models. In addition, MAOP learns semantically and visually interpretable disentangled representations. |
| Researcher Affiliation | Academia | 1Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing, China 2Department of Computer Science and Technology, Tsinghua University, Beijing, China {zhu-gx15, wjh19, rzz16, linzc16}@mails.tsinghua.edu.cn, chongjie@tsinghua.edu.cn |
| Pseudocode | Yes | Algorithm 1 shows a pseudocode that summarizes the training process of our framework. |
| Open Source Code | No | The paper repeatedly references 'Supplementary Material (https://arxiv.org/abs/1904.07482)' for algorithms, implementation details, figures, and videos, but it does not contain an explicit statement that the source code for the methodology is released or provide a direct link to a code repository. |
| Open Datasets | Yes | We first evaluate zero-shot generalization and sample efficiency on Monster Kong from Pygame Learning Environment (Tasfi2016), which allows us to test generalization ability over various scenes with different layouts. |
| Dataset Splits | No | The paper states 'trained with 3k samples from a single environment' and provides training sample counts like '3k', '100k', '200k', '300k', and '500k'. It uses 'training environments' and 'unseen environments for testing' but does not explicitly describe a separate validation dataset split (e.g., 'X% for validation'). |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions 'Pygame Learning Environment' and 'Mask R-CNN' as methods/environments, but it does not list any specific software components with their version numbers that would be necessary to reproduce the experiment. |
| Experiment Setup | Yes | MAOP employs a three-level learning architecture from the most abstract level of motion detection, to dynamic instance segmentation, and to dynamics learning and prediction. The whole architecture includes four major components: A) an Object Detector that learns to decompose the input image into objects; B) an Instance Localization module responsible for localizing dynamic instances; C) a Dynamics Net for learning the motion of each dynamic instance conditioned on the effects from actions and object-level spatial-temporal relations; and D) a Background Constructor that computes a background image from learned static object masks. The total loss of this level is given by, LDIS = Linstance + λ3Lmerge + λ4Lforground. Therefore, the total loss of the dynamics learning level is given by, LDL = Lpred-object + λ1Lpred-image + λ2Lproposal. h is the history length. |