Reproduce “Adaptive Wiping: Adaptive contact-rich manipulation through few-shot imitation learning with Force-Torque feedback and pre-trained object representations” 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
Chikaha Tsuji, Enrique Coronado, Pablo Osorio, Gentiane Venture
A robotic framework enables adaptive contact-rich wiping of deformable objects by integrating few-shot imitation learning with real-time Force-Torque feedback and pre-trained object representations. The system achieved 100% contact and applied 96% of the human-demonstrated reference force across unseen surface heights and varied sponge properties, demonstrating robust adaptation.
Key insights
- Pre-training a VAE on exploratory force-torque data effectively extracts a low-dimensional latent representation of deformable object properties, enabling generalization to unseen objects.
- Integrating real-time force-torque feedback into a closed-loop control framework allows robots to adapt their vertical motion to dynamic environmental changes with minimal demonstrations.
- A two-step learning approach, separating object property representation from task learning, enhances data efficiency and adaptability for contact-rich tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.