Summary of Intrinsic Dynamics-driven Generalizable Scene Representations For Vision-oriented Decision-making Applications, by Dayang Liang et al.
Intrinsic Dynamics-Driven Generalizable Scene Representations for Vision-Oriented Decision-Making Applications
by Dayang Liang, Jinyang Lai, Yunlong Liu
First submitted to arxiv on: 30 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes an intrinsic dynamics-driven representation learning method, called DSR, to improve the ability of scene representation in vision-oriented decision-making applications. Current approaches typically learn task-relevant state representations within visual reinforcement learning, but often ignore inherent dynamics relationships among elements. The proposed method optimizes a parameterized encoder by the state-transition dynamics of the underlying system, prompting latent encoding information to satisfy state transitions and distinguish between state and noise spaces. To further improve representation ability, sequential elements’ frequency domain and multi-step prediction are adopted for sequentially modeling inherent dynamics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper aims to solve the problem of learning accurate scene representations in vision-oriented decision-making applications. It proposes a new method called DSR that takes into account the dynamics relationships among elements, which helps to distinguish between state and noise spaces. The method is tested on visual distracting DMControl control tasks and achieves significant performance improvements over the backbone baseline. The paper also shows that DSR can be used in real-world autonomous driving applications on the CARLA simulator. |
Keywords
» Artificial intelligence » Encoder » Prompting » Reinforcement learning » Representation learning
