DRAE: Dynamic Retrieval-Augmented Expert Networks for Lifelong Learning and Task Adaptation in Robotics
Yayu Long, Kewei Chen, Long Jin, Mingsheng Shang
TL;DR
DRAE tackles lifelong learning and catastrophic forgetting in robotic control by unifying dynamic Mixture-of-Experts routing with parameterized Retrieval-Augmented Generation and a hierarchical Reinforcement Learning framework coordinated by ReflexNet-SchemaPlanner-HyperOptima (RSHO). It introduces a non-parametric memory module (DPMM) to preserve older skills while enabling continual growth, and a unified objective that adaptively balances short-term adaptation with long-term knowledge retention. The approach yields state-of-the-art results across diverse robotic benchmarks ( MimicGen, DexArt, Adroit, NAVSIM) and shows strong generalization, reduced forgetting, and competitive real-world deployment performance, with theoretical guarantees on dynamic regret and sample complexity. By enabling dynamic expert expansion guided by external knowledge, DRAE provides a scalable, memory-efficient foundation for lifelong learning in dynamic robotic environments, with practical impact on manipulation, navigation, and humanoid motion tasks. Key results include an 82.5% task success rate on dynamic robotic manipulation tasks (vs. 74.2% for static MoEs), and substantial improvements in real-world deployment metrics such as higher success rates and faster adaptation when transferring to unseen domains.
Abstract
We introduce Dynamic Retrieval-Augmented Expert Networks (DRAE), a groundbreaking architecture that addresses the challenges of lifelong learning, catastrophic forgetting, and task adaptation by combining the dynamic routing capabilities of Mixture-of-Experts (MoE); leveraging the knowledge-enhancement power of Retrieval-Augmented Generation (RAG); incorporating a novel hierarchical reinforcement learning (RL) framework; and coordinating through ReflexNet-SchemaPlanner-HyperOptima (RSHO).DRAE dynamically routes expert models via a sparse MoE gating mechanism, enabling efficient resource allocation while leveraging external knowledge through parametric retrieval (P-RAG) to augment the learning process. We propose a new RL framework with ReflexNet for low-level task execution, SchemaPlanner for symbolic reasoning, and HyperOptima for long-term context modeling, ensuring continuous adaptation and memory retention. Experimental results show that DRAE significantly outperforms baseline approaches in long-term task retention and knowledge reuse, achieving an average task success rate of 82.5% across a set of dynamic robotic manipulation tasks, compared to 74.2% for traditional MoE models. Furthermore, DRAE maintains an extremely low forgetting rate, outperforming state-of-the-art methods in catastrophic forgetting mitigation. These results demonstrate the effectiveness of our approach in enabling flexible, scalable, and efficient lifelong learning for robotics.