Summary of Moto: Offline Pre-training to Online Fine-tuning For Model-based Robot Learning, by Rafael Rafailov et al.
MOTO: Offline Pre-training to Online Fine-tuning for Model-based Robot Learning
by Rafael Rafailov, Kyle Hatch, Victor Kolev, John D. Martin, Mariano Phielipp, Chelsea Finn
First submitted to arxiv on: 6 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 approach, MOTO, combines model-based reinforcement learning with offline pre-training and online fine-tuning for realistic robot tasks. It tackles issues like distribution shifts, off-dynamics data, and non-stationary rewards by leveraging epistemic uncertainty control and policy regularization. MOTO efficiently reuses prior data through model-based value expansion while preventing model exploitation. The method successfully solves tasks from the MetaWorld benchmark and Franka Kitchen environment, completely from images. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Moto is a new way to make robots learn and adapt to new situations. It helps them use what they learned before to solve new problems. Moto fixes some problems with old methods that don’t work well in high-dimensional domains. It makes robots better at solving tasks by using prior knowledge and controlling how sure it is about its predictions. |
Keywords
* Artificial intelligence * Fine tuning * Regularization * Reinforcement learning
