mlsanityData to 14 Oct 2026

Robotic Policy Adaptation via Weight-Space Meta-Learning

arXiv 2606.07217 ↗First posted 5 Jun 2026cs.RO cs.CV cs.LGcited by 1

On the map: Vision-language-action models → Reinforcement learning and post-training

Abstract (arXiv)

Vision-Language-Action (VLA) models are emerging as a promising paradigm for robotic manipulation, enabling general-purpose policies trained from large corpora of demonstrations and action labels. However, adapting these models to new tasks still typically requires task-specific demonstrations, action annotations, and additional fine-tuning, making deployment costly and difficult to scale. We propose WIZARD, a weight-space meta-learning framework that sidesteps task-specific fine-tuning by generating task-specific LoRA parameters for a frozen VLA policy. Given only a language instruction and a short demonstration video, WIZARD predicts the corresponding adaptation weights in a single forward pass, without target-task action labels or test-time optimization. During meta-training, WIZARD learns to map task evidence directly to expert LoRA updates, capturing relationships between tasks in weight space. Experiments on LIBERO show that WIZARD improves performance by up to ~2x on unseen dataset collections and up to ~14x on unseen tasks. On a Franka Emika Panda, WIZARD consistently improves over a real-domain adapted baseline, showing that generated adapters provide task-level specialization beyond simulation.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Reinforcement learning and post-training, one of 13 directions in Vision-language-action models. Builds on 29 of its 35 references that are in the map. First posted on arXiv 5 Jun 2026.

Builds on

29 of 35 references.

References · 29

  1. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,887
  2. 2023PaLM-E: An Embodied Multimodal Language Modelcited by 2,071
  3. 2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,061
  4. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,761
  5. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,410
  6. 2021RMA: Rapid Motor Adaptation for Legged Robotscited by 508
  7. 2022VIMA: General Robot Manipulation with Multimodal Promptscited by 407
  8. 2021MT-Opt: Continuous Multi-Task Robotic Reinforcement Learning at Scalecited by 229
  9. 2019Task2Vec: Task Embedding for Meta-Learningcited by 205
  10. 2017SMASH: One-Shot Model Architecture Search through HyperNetworkscited by 202
  11. 2025LIBERO-PRO: Towards Robust and Fair Evaluation of Vision-Language-Action Models Beyond Memorizationcited by 179
  12. 2016Using Fast Weights to Attend to the Recent Pastcited by 170
17 more
  1. 2021Parameter Prediction for Unseen Deep Architecturescited by 89
  2. 2022Learning to Learn with Generative Models of Neural Network Checkpointscited by 86
  3. 2024Prediction with Action: Visual Policy Learning via Joint Denoising Processcited by 70
  4. 2018Task-Embedded Control Networks for Few-Shot Imitation Learningcited by 66
  5. 2022Hyper-Representations as Generative Models: Sampling Unseen Neural Network Weightscited by 64
  6. 2022Hypernetworks in Meta-Reinforcement Learningcited by 42
  7. 2024Diffusion-Based Neural Network Weights Generationcited by 39
  8. 2025METAL: A Multi-Agent Framework for Chart Generation with Test-Time Scalingcited by 34
  9. 2025Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weightscited by 20
  10. 2024One-Shot Imitation Learning with Invariance Matching for Robotic Manipulationcited by 18
  11. 2024Conditional LoRA Parameter Generationcited by 16
  12. 2025Recurrent Diffusion for Large-Scale Parameter Generationcited by 15
  13. 2024Make-An-Agent: A Generalizable Policy Network Generator with Behavior-Prompted Diffusioncited by 14
  14. 2025MetaVLA: Unified Meta Co-training For Efficient Embodied Adaptioncited by 5
  15. 2024WARPD: World model Assisted Reactive Policy Diffusioncited by 4
  16. 2025Hyper-GoalNet: Goal-Conditioned Manipulation Policy Learning with HyperNetworkscited by 4
  17. 2013Generative NeuroEvolution for Deep Learningcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Amortized Low-Rank Adaptation for Model-Based Reinforcement Learningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Policy Adaptation via Language Optimization: Decomposing Tasks for Few-Shot Imitation
  2. 2025MoS-VLA: A Vision-Language-Action Model with One-Shot Skill Adaptation
  3. 2026RA-VLA: Retrieval-Augmented VLA for Test-Time Adaptation
  4. 2026Agentic-VLA: Efficient Online Adaptation for Vision-Language-Action Models
  5. 2026RARA: Robust Policy Fine-Tuning through Retrieval-Anchored Representation Alignment for Robotic Manipulation
  6. 2026AtomVLA: Scalable Post-Training for Robotic Manipulation via Predictive Latent World Models
  7. 2026PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models
  8. 2026Inline Memory Meets Reusable Skills: Memory-centric Framework for Vision-Language-Action Model