FLAME: A Federated Learning Benchmark for Robotic Manipulation
Santiago Bou Betran, Alberta Longhini, Miguel Vasco, Yuchong Zhang, Danica Kragic
2025-03-06
Summary
This paper talks about FLAME, a new way to test how robots can learn to manipulate objects by sharing knowledge without sharing private data
What's the problem?
Currently, robots learn to handle objects using big datasets stored in one place. This method doesn't protect privacy well and makes it hard for robots to adapt to new situations quickly
What's the solution?
The researchers created FLAME, which includes over 160,000 examples of robots doing tasks in different virtual environments. They also made a system to test how well robots can learn from each other without sharing raw data. This method, called federated learning, lets robots improve their skills while keeping their experiences private
Why it matters?
This matters because it could help robots learn faster and work better in different places without risking privacy. It's like robots sharing tips without telling each other everything they've done. This could lead to smarter, more adaptable robots that can be used in more places, like homes or hospitals, where privacy is important
Abstract
Recent progress in robotic manipulation has been fueled by large-scale datasets collected across diverse environments. Training robotic manipulation policies on these datasets is traditionally performed in a centralized manner, raising concerns regarding scalability, adaptability, and data privacy. While federated learning enables decentralized, privacy-preserving training, its application to robotic manipulation remains largely unexplored. We introduce FLAME (Federated Learning Across Manipulation Environments), the first benchmark designed for federated learning in robotic manipulation. FLAME consists of: (i) a set of large-scale datasets of over 160,000 expert demonstrations of multiple manipulation tasks, collected across a wide range of simulated environments; (ii) a training and evaluation framework for robotic policy learning in a federated setting. We evaluate standard federated learning algorithms in FLAME, showing their potential for distributed policy learning and highlighting key challenges. Our benchmark establishes a foundation for scalable, adaptive, and privacy-aware robotic learning.