Research map / Detection and segmentation / Open-set and few-shot detection
Object-Oriented Dynamics Learning through Multi-Level Abstraction
Guangxiang Zhu, Jianhao Wang, Zhizhou Ren, Zichuan Lin, Chongjie Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use the three-level MAOP framework for object-oriented dynamics learning | action-conditioned dynamics prediction · Monster Kong, 1-5, 3k samples, unseen environments | 0-error accuracy: 0.14 -> 0.94 | Improved |
| use MAOP instead of CDNA | action-conditioned dynamics prediction · Flappy Bird, 100 samples, Agent | 0-acc: 0.13 -> 0.83 | Improved |
| use MAOP instead of CDNA | action-conditioned dynamics prediction · Freeway, 100 samples | 0-acc: 0.42 -> 0.78 | Improved |
| use MAOP instead of CDNA | action-conditioned dynamics prediction · Krull, 3k samples | 0acc: 0.27 -> 0.14 | Hurt |
| use MAOP instead of OODP | action-conditioned dynamics prediction · Breakout, 3k samples | 0acc: 0.73 -> 0.52 | Hurt |
| use MCTS with the learned MAOP dynamics model instead of DQN | planning in unseen Monster Kong environments · Monster Kong | Reward: 13.67 -> 38.19 | Improved |
| use the learned MAOP model rather than the real simulator for MCTS | planning in unseen Monster Kong environments · Monster Kong | Reward: 38.41 -> 38.19 | No clear effect |
| replace DIS instance proposals with Gaussian-noise proposals (ablation) | MAOP dynamics learning · Monster Kong, unseen environments, 3-to-5 generalization, Agent | 0-acc: 0.99 -> 0.60 | Hurt |
| replace DIS instance proposals with Poisson-noise proposals (ablation) | MAOP dynamics learning · Monster Kong, unseen environments, 3-to-5 generalization, Agent | 0-acc: 0.99 -> 0.93 | Hurt |
| use 3 rather than 5 object masks (ablation) | MAOP · Monster Kong, unseen environments, Agent | 0-error accuracy: 0.94 -> 0.90 | Hurt |
| use 8 rather than 5 object masks (ablation) | MAOP · Monster Kong, unseen environments, Agent | 0-error accuracy: 0.94 -> 0.91 | Hurt |
| train with 3k rather than 300k samples | MAOP versus CDNA · Monster Kong, unseen environments, Agent | 0-error accuracy: CDNA: 300k samples, 0.26; MAOP: 3k samples, 0.94 | Improved |
Closest papers by meaning
- Object-Oriented Dynamics Predictor (NeurIPS 2018 · similarity 0.91)
- HyperDynamics: Meta-Learning Object and Agent Dynamics with Hypernetworks (ICLR 2021 · similarity 0.86)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.85)
- Learning Causal Dynamics Models in Object-Oriented Environments (ICML 2024 · similarity 0.85)
- DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control (NeurIPS 2024 · similarity 0.85)
- Seek Commonality but Preserve Differences: Dissected Dynamics Modeling for Multi-modal Visual RL (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
