RoboFL: Federated Expert Assembly for World Action Models

2026-09-28

AI Summary

This study addresses the challenges of adapting federated learning to world action models, where physical interaction data are scarce, siloed, and task-heterogeneous. To this end, we propose MoSAIC, an architecture that formulates local LoRA adapters as expert branches and introduces a novel hybrid slot adapter mounting strategy. By integrating Forward-to-Action Routing Distillation (FARD) with Path Consensus Expert Aggregation (PCEA), MoSAIC effectively prevents update entanglement while preserving expert specificity. Experiments on Franka robotic manipulators demonstrate that our method surpasses the centralized baseline by 12.23% and reduces communication overhead by 86.81% compared to Mixture-of-Experts baselines, thereby achieving efficient federated world action learning.

Abstract

Vision-language-action and world-action models are increasingly popular, yet remain bottlenecked by physical interaction data that is scarce, institutionally siloed, and task-heterogeneous. A natural federated solution is to let each client adapt a shared foundation model through parameter-efficient fine-tuning, avoiding the exchange of full-model updates. However, federating these adapters is nontrivial, as naive aggregation can entangle incompatible updates, while incorporating MoE-style routing into federated aggregation may dilute specialization and destabilize expert selection. We present RoboFL, which instantiates MoSAIC (Mixture of Slotted Adapters) for federated world-action learning. MoSAIC directly installs locally trained LoRA adapters as the expert branches of a server MoE. Server-side routers learn token assignments over these prior-informed branches while jointly refining routing and expert parameters. Foresight-to-Action Routing Distillation (FARD) aligns routing across the model's three paths, while Path-Consensus Expert Aggregation (PCEA) converts complete expert updates into a compact global adapter for personalized redistribution. Experiments on RoboTwin 2.0, RLBench, and a real-world Franka robot arm show the superiority of RoboFL with structured expert assembly, as it outperforms centralized PEFT InternVLA-A1 by 12.23% on the Franka arm, while reducing per-round client communication by up to 86.81% relative to MoE-based federated VLA baselines.

Problem

• Federated Learning
• Vision-Language-Action Models
• World Action Models
• Parameter-Efficient Fine-Tuning
• Adapter Aggregation

Innovation

• Federated Learning
• Mixture of Experts
• LoRA Adapters
• World Action Models
• Routing Distillation