Summary of Off-dynamics Conditional Diffusion Planners, by Wen Zheng Terence Ng et al.
Off-dynamics Conditional Diffusion Planners
by Wen Zheng Terence Ng, Jianda Chen, Tianwei Zhang
First submitted to arxiv on: 16 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Robotics (cs.RO)
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 proposed novel approach uses conditional Diffusion Probabilistic Models (DPMs) to learn the joint distribution of off-dynamics datasets and limited target datasets. This enables Offline Reinforcement Learning (RL) to address data scarcity challenges. The model captures underlying dynamics structure using two contexts: continuous dynamics scores for partial overlap and inverse-dynamics contexts guiding trajectory generation adhering to target environment constraints. Results show significant performance improvements over strong baselines, with ablation studies highlighting the critical role of each context. Additionally, the model demonstrates robustness to subtle environmental shifts by modifying its context. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Offline RL uses pre-existing datasets instead of interactive data acquisition. This paper explores using off-dynamics datasets to address data scarcity issues in Offline RL. The method learns a joint distribution of large-scale off-dynamics datasets and limited target datasets using conditional DPMs. Two contexts help the model capture underlying dynamics: continuous dynamics scores for partial overlap and inverse-dynamics contexts guiding trajectory generation. Results show better performance than strong baselines, with ablation studies highlighting context importance. |
Keywords
» Artificial intelligence » Diffusion » Reinforcement learning
