arXiv cs.AI· Fanqi Yu, Matteo Tiezzi, Cigdem Beyan, Tommaso Apicella, Vittorio Murino·· 13 小时前AI 评分31
HIDRA：面向无回放终身模仿学习的分层双路由注意力机制
HIDRA: Hierarchical Dual-Routing Attention for Replay-Free Lifelong Imitation Learning
AI 导读
HIDRA 提出分层双注意力路由框架，将指令级专家检索与上下文相关细化解耦，缓解无回放单遍终身模仿学习中的路由干扰问题。该方法引入 key 级几何正则化，在任务内对齐、任务间分离，以稳定表征漂移下的路由。在 LIBERO 多个基准（含异构任务序列与改写语言指令变体）上，HIDRA 的 AUC 与前向迁移均优于无回放和有回放基线，且保持低遗忘。
正文 · 原文
Abstract:Robotic agents operating in real-world environments must adapt continuously from a stream of multimodal demonstrations, often under constraints that preclude storing or revisiting past data. In this replay-free, single-pass lifelong imitation learning setting, sequential updates can progressively distort the representation space, making expert-based approaches particularly sensitive to routing interference, causing expert selection to degrade over time. We propose HIDRA, a structured routing framework that mitigates this issue through a hierarchical dual-attention mechanism, which decouples instruction-level expert retrieval from context-dependent refinement. To further stabilize routing under representation drift, HIDRA introduces a key-level geometric regularization that enforces both alignment within tasks and separation across tasks. The resulting approach enables reliable expert selection while supporting an expandable set of experts for continual adaptation. We evaluate HIDRA on several LIBERO benchmark suites, including more challenging variants with heterogeneous task sequences and paraphrased language instructions. Our approach consistently outperforms both replay-free and replay-based baselines, improving AUC and forward transfer while maintaining low forgetting, with stronger gains under high interference and language variation. The code is available at: this https URL.
| Comments: | This is the authors' preprint, accepted to the NeurIPS 2026 Conference. The final version is available in the NeurIPS 2026 proceedings |
| Subjects: | Robotics (cs.RO); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.36374 [cs.RO] |
| (or arXiv:2610.36374v1 [cs.RO] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36374 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Cigdem Beyan [view email]
[v1]
Mon, 26 Oct 2026 22:43:30 UTC (973 KB)
来源：arXiv cs.AI · arxiv.org