# Actions as Language: Fine-Tuning VLMs into VLAs Without Catastrophic Forgetting

arXiv: https://arxiv.org/abs/2509.22195 · HTML: https://mlsanity.com/paper/2509.22195/

## Key facts

- Cited by 58 papers in the map (data through 14 Oct 2026).
- Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning.
- Builds on 17 of its 26 references that are in the map.
- First posted on arXiv 26 Sep 2025.

## Placed in

- LLM reasoning / [Spatial and embodied reasoning](https://mlsanity.com/lr/spatial-and-embodied-reasoning/)
- Multimodal LLMs / vision-language models / [Embodied and 3D vision-language models](https://mlsanity.com/mllm/embodied-and-3d-vision-language-models/)
- Vision-language-action models / [Generalist VLA models and benchmarks](https://mlsanity.com/vla/generalist-vla-models-and-benchmarks/)

## Builds on (most cited first, 17 of 17)

- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control](https://mlsanity.com/paper/2307.15818/) (2023)
- [RT-1: Robotics Transformer for Real-World Control at Scale](https://mlsanity.com/paper/2212.06817/) (2022)
- [π0.5: a Vision-Language-Action Model with Open-World Generalization](https://mlsanity.com/paper/2504.16054/) (2025)
- [GR00T N1: An Open Foundation Model for Generalist Humanoid Robots](https://mlsanity.com/paper/2503.14734/) (2025)
- [Open X-Embodiment: Robotic Learning Datasets and RT-X Models](https://mlsanity.com/paper/2310.08864/) (2023)
- [SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Model](https://mlsanity.com/paper/2501.15830/) (2025)
- [CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models](https://mlsanity.com/paper/2503.22020/) (2025)
- [Generalized Planning in PDDL Domains with Pretrained Large Language Models](https://mlsanity.com/paper/2305.11014/) (2023)
- [Yell At Your Robot: Improving On-the-Fly from Language Corrections](https://mlsanity.com/paper/2403.12910/) (2024)
- [OneTwoVLA: A Unified Vision-Language-Action Model with Adaptive Reasoning](https://mlsanity.com/paper/2505.11917/) (2025)
- [LLARVA: Vision-Action Instruction Tuning Enhances Robot Learning](https://mlsanity.com/paper/2406.11815/) (2024)
- [ChatVLA-2: Vision-Language-Action Model with Open-World Embodied Reasoning from Pretrained Knowledge](https://mlsanity.com/paper/2505.21906/) (2025)
- [Run-time Observation Interventions Make Vision-Language-Action Models More Visually Robust](https://mlsanity.com/paper/2410.01971/) (2024)
- [Enhancing Generalization in Vision-Language-Action Models by Preserving Pretrained Representations](https://mlsanity.com/paper/2509.11417/) (2025)
- [Robix: A Unified Model for Robot Interaction, Reasoning and Planning](https://mlsanity.com/paper/2509.01106/) (2025)
- [Is Your Imitation Learning Policy Better than Mine? Policy Comparison with Near-Optimal Stopping](https://mlsanity.com/paper/2503.10966/) (2025)

## Built on by (most cited first, 30 of 58)

- [SIMA 2: A Generalist Embodied Agent for Virtual Worlds](https://mlsanity.com/paper/2512.04797/) (2025)
- [ReMem-VLA: Empowering Vision-Language-Action Model with Memory via Dual-Level Recurrent Queries](https://mlsanity.com/paper/2603.12942/) (2026)
- [LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queries](https://mlsanity.com/paper/2601.15197/) (2026)
- [In-N-On: Scaling Egocentric Manipulation with in-the-wild and on-task Data](https://mlsanity.com/paper/2511.15704/) (2025)
- [FASTer: Toward Efficient Autoregressive Vision Language Action Modeling via Neural Action Tokenization](https://mlsanity.com/paper/2512.04952/) (2025)
- [A Systematic Study of Data Modalities and Strategies for Co-training Large Behavior Models for Robot Manipulation](https://mlsanity.com/paper/2602.01067/) (2026)
- [Scaling Verification Can Be More Effective than Scaling Policy Learning for Vision-Language-Action Alignment](https://mlsanity.com/paper/2602.12281/) (2026)
- [MergeVLA: Cross-Skill Model Merging Toward a Generalist Vision-Language-Action Agent](https://mlsanity.com/paper/2511.18810/) (2025)
- [LAP: Language-Action Pre-Training Enables Zero-shot Cross-Embodiment Transfer](https://mlsanity.com/paper/2602.10556/) (2026)
- [ActionCodec: What Makes for Good Action Tokenizers](https://mlsanity.com/paper/2602.15397/) (2026)
- [Scaling World Model for Hierarchical Manipulation Policies](https://mlsanity.com/paper/2602.10983/) (2026)
- [TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformers](https://mlsanity.com/paper/2601.14133/) (2026)
- [Robotic Video World Models: A Survey of Applications, Research Challenges, Future Directions](https://mlsanity.com/paper/2601.07823/) (2026)
- [VLA-Trace: Diagnosing Vision-Language-Action Models through Representation and Behavior Tracing](https://mlsanity.com/paper/2605.30117/) (2026)
- [Towards Long-horizon Embodied Agents with Tool-Aligned Vision-Language-Action Models](https://mlsanity.com/paper/2605.13119/) (2026)
- [HiVLA: A Visual-Grounded-Centric Hierarchical Embodied Manipulation System](https://mlsanity.com/paper/2604.14125/) (2026)
- [MAPS: Preserving Vision-Language Representations via Module-Wise Proximity Scheduling for Better Vision-Language-Action Generalization](https://mlsanity.com/paper/2511.19878/) (2025)
- [CRL-VLA: Continual Vision-Language-Action Learning](https://mlsanity.com/paper/2602.03445/) (2026)
- [How Do VLAs Effectively Inherit from VLMs?](https://mlsanity.com/paper/2511.06619/) (2025)
- [Generalizable VLA Finetuning via Representation Anchoring and Language-Action Alignment](https://mlsanity.com/paper/2607.13429/) (2026)
- [ReViP: Mitigating False Completion in Vision-Language-Action Models with Vision-Proprioception Rebalance](https://mlsanity.com/paper/2601.16667/) (2026)
- [Veo-Act: Enhancing VLA Policies with Frontier Video Models](https://mlsanity.com/paper/2604.04502/) (2026)
- [VIA: Visual Interface Agent for Robot Control](https://mlsanity.com/paper/2607.11119/) (2026)
- [NoTVLA: Semantics-Preserving Robot Adaptation via Narrative Action Interfaces](https://mlsanity.com/paper/2510.03895/) (2025)
- [TORL-VLA: Tactile Guided Online Reinforcement Learning for Contact-Rich Manipulation](https://mlsanity.com/paper/2606.09337/) (2026)
- [Look Before You Leap: Distilling Tree Search into Action Evaluation for Frozen VLA Models](https://mlsanity.com/paper/2607.03751/) (2026)
- [StellaVLA: In-Context Structured Demonstration for Generalizable Vision-Language-Action Models](https://mlsanity.com/paper/2608.11671/) (2026)
- [VISTA: Enhancing Visual Conditioning via Track-Following Preference Optimization in Vision-Language-Action Models](https://mlsanity.com/paper/2602.05049/) (2026)
- [Robo-Harness K1: Harnessing Robot-Use Agents via Perception Augmentation](https://mlsanity.com/paper/2609.29389/) (2026)
- [VLA-GSE: Boosting Parameter-Efficient Fine-Tuning in VLA with Generalized and Specialized Experts](https://mlsanity.com/paper/2605.06175/) (2026)

## Abstract (arXiv)

Fine-tuning vision-language models (VLMs) on robot teleoperation data to create vision-language-action (VLA) models is a promising paradigm for training generalist policies, but it suffers from a fundamental tradeoff: learning to produce actions often diminishes the VLM's foundational reasoning and multimodal understanding, hindering generalization to novel scenarios, instruction following, and semantic understanding. We argue that this catastrophic forgetting is due to a distribution mismatch between the VLM's internet-scale pretraining corpus and the robotics fine-tuning data. Inspired by this observation, we introduce VLM2VLA: a VLA training paradigm that first resolves this mismatch at the data level by representing low-level actions with natural language. This alignment makes it possible to train VLAs solely with Low-Rank Adaptation (LoRA), thereby minimally modifying the VLM backbone and averting catastrophic forgetting. As a result, the VLM can be fine-tuned on robot teleoperation data without fundamentally altering the underlying architecture and without expensive co-training on internet-scale VLM datasets. Through extensive Visual Question Answering (VQA) studies and over 800 real-world robotics experiments, we demonstrate that VLM2VLA preserves the VLM's core capabilities, enabling zero-shot generalization to novel tasks that require open-world semantic reasoning and multilingual instruction following.