Reproduce “Interactive Imitation Learning in Robotics: A Survey” 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
Carlos Celemin, Rodrigo Pérez-Dattari, Eugenio Chisari, Giovanni Franzese, Leandro de Souza Rosa, Ravi Prakash, Zlatan Ajanović, Marta Ferraz and 2 more
A survey unifies Interactive Imitation Learning (IIL) in robotics, establishing a clear terminology and a comprehensive classification framework for methods, human-robot interfaces, and applications. The work also identifies critical challenges and outlines future research directions for developing more intuitive and robust robot learning systems.
Key insights
- IIL offers a data-efficient and robust alternative to traditional IL and RL, inherently minimizing distribution mismatch by allowing human feedback directly on the robot's own executed trajectories.
- Human feedback in IIL is highly diverse, categorized into reinforcements, preferences, and absolute/relative corrections, each offering different levels of information richness versus human effort.
- Auxiliary models are crucial for enhancing IIL by supporting dimensionality reduction, generalization, planning, active querying, and effectively interpreting human feedback and intentions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.