Reproduce “MILES: Making Imitation Learning Easy with Self-Supervision” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Georgios Papagiannis, Edward Johns
MILES enables robots to learn complex manipulation tasks from a single human demonstration and one environment reset by autonomously augmenting the training data. This framework achieves success rates between 70% and 95% across seven diverse real-world contact-rich tasks like USB insertion and key manipulation.
Key insights
- RGB vision is the most critical modality for MILES's performance, with force feedback providing additional, but secondary, benefits.
- The required amount of self-supervised data varies with task complexity, with precision tasks like USB insertion benefiting more from extensive data augmentation.
- Performance generally improves with larger augmentation datasets, but the rate of improvement and the optimal dataset size can vary significantly depending on the specific task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.