Achieves 45% performance gains in robotics using 5-10x fewer real-world demonstrations through high-dimensional factorization.
March 27, 2026
Original Paper
Towards Generalizable Robotic Data Flywheel: High-Dimensional Factorization and Composition
arXiv · 2603.25583
The Takeaway
It introduces a structured 'data flywheel' strategy that decomposes tasks into object, action, and environment factors. This allows for compositional generalization, solving the primary bottleneck in generalist robotic learning.
From the abstract
The lack of sufficiently diverse data, coupled with limited data efficiency, remains a major bottleneck for generalist robotic models, yet systematic strategies for collecting and curating such data are not fully explored. Task diversity arises from implicit factors that are sparsely distributed across multiple dimensions and are difficult to define explicitly. To address this challenge, we propose F-ACIL, a heuristic factor-aware compositional iterative learning framework that enables structure