Summary of Whale: Towards Generalizable and Scalable World Models For Embodied Decision-making, by Zhilong Zhang et al.
WHALE: Towards Generalizable and Scalable World Models for Embodied Decision-making
by Zhilong Zhang, Ruifeng Chen, Junyin Ye, Yihao Sun, Pengyuan Wang, Jingcheng Pang, Kaiyuan Li, Tianshuo Liu, Haoxin Lin, Yang Yu, Zhi-Hua Zhou
First submitted to arxiv on: 8 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 introduces WHALE, a framework for learning generalizable world models, crucial for decision-making in embodied environments. It combines two key techniques: behavior-conditioning to address policy distribution shifts, and retracing-rollout for efficient uncertainty estimation without model ensembles. The framework is universal and can be combined with any neural network architecture. The paper presents Whale-ST, a scalable spatial-temporal transformer-based world model with enhanced generalizability. Experimental results demonstrate the superiority of Whale-ST in simulation tasks, and the effectiveness of its uncertainty estimation technique enhances model-based policy optimization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary World models help robots make decisions by letting them try out different actions without actually doing them. To make good decisions, these models need to be very good at imagining what might happen if they do something. They also need to be able to guess how sure they are about their predictions. This paper introduces a new way to learn world models that can do both of these things well. It uses two techniques: one helps the model imagine different scenarios, and the other helps it make good guesses about its own uncertainty. The authors show that this approach works better than previous methods in simulation tasks and in real-world situations where robots need to use minimal demonstrations. |
Keywords
» Artificial intelligence » Neural network » Optimization » Transformer
