# Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models

arXiv: https://arxiv.org/abs/2603.19183 · HTML: https://mlsanity.com/paper/2603.19183/

## Key facts

- Cited by 16 papers in the map (data through 14 Oct 2026).
- Placed in Generalist VLA models and benchmarks, one of 13 directions in Vision-language-action models.
- Builds on 22 of its 30 references that are in the map.
- First posted on arXiv 19 Mar 2026.

## Placed in

- Vision-language-action models / [Generalist VLA models and benchmarks](https://mlsanity.com/vla/generalist-vla-models-and-benchmarks/)

## Builds on (most cited first, 22 of 22)

- [Visual Instruction Tuning](https://mlsanity.com/paper/2304.08485/) (2023)
- [OpenVLA: An Open-Source Vision-Language-Action Model](https://mlsanity.com/paper/2406.09246/) (2024)
- [π0: A Vision-Language-Action Flow Model for General Robot Control](https://mlsanity.com/paper/2410.24164/) (2024)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [π0.5: a Vision-Language-Action Model with Open-World Generalization](https://mlsanity.com/paper/2504.16054/) (2025)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning](https://mlsanity.com/paper/2306.03310/) (2023)
- [Open X-Embodiment: Robotic Learning Datasets and RT-X Models](https://mlsanity.com/paper/2310.08864/) (2023)
- [DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset](https://mlsanity.com/paper/2403.12945/) (2024)
- [PaliGemma: A versatile 3B VLM for transfer](https://mlsanity.com/paper/2407.07726/) (2024)
- [Scaling and evaluating sparse autoencoders](https://mlsanity.com/paper/2406.04093/) (2024)
- [RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots](https://mlsanity.com/paper/2406.02523/) (2024)
- [LIBERO-PRO: Towards Robust and Fair Evaluation of Vision-Language-Action Models Beyond Memorization](https://mlsanity.com/paper/2510.03827/) (2025)
- [Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better](https://mlsanity.com/paper/2505.23705/) (2025)
- [Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models](https://mlsanity.com/paper/2504.02821/) (2025)
- [Building Production-Ready Probes For Gemini](https://mlsanity.com/paper/2601.11516/) (2026)
- [Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization](https://mlsanity.com/paper/2510.25616/) (2025)
- [Mechanistic interpretability for steering vision-language-action models](https://mlsanity.com/paper/2509.00328/) (2025)
- [Enhancing Generalization in Vision-Language-Action Models by Preserving Pretrained Representations](https://mlsanity.com/paper/2509.11417/) (2025)
- [Probing a Vision-Language-Action Model for Symbolic States and Integration into a Cognitive Architecture](https://mlsanity.com/paper/2502.04558/) (2025)
- [Observing and Controlling Features in Vision-Language-Action Models](https://mlsanity.com/paper/2603.05487/) (2026)
- [Emergent World Representations in OpenVLA](https://mlsanity.com/paper/2509.24559/) (2025)

## Built on by (most cited first, 16 of 16)

- [VLA-Trace: Diagnosing Vision-Language-Action Models through Representation and Behavior Tracing](https://mlsanity.com/paper/2605.30117/) (2026)
- [Point Tracking Improves World Action Models](https://mlsanity.com/paper/2605.23856/) (2026)
- [Learning What to Say to Your VLA: Mostly Harmless Vision Language Action Model Steering](https://mlsanity.com/paper/2606.12299/) (2026)
- [Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control](https://mlsanity.com/paper/2607.14943/) (2026)
- [Decoding Task Progress from VLA Representations](https://mlsanity.com/paper/2608.13474/) (2026)
- [Event-Grounded Sparse Autoencoders for Vision-Language-Action Policies](https://mlsanity.com/paper/2605.17204/) (2026)
- [Sequential Planning via Anchored Robotic Keypoints](https://mlsanity.com/paper/2606.30613/) (2026)
- [Mechanistic Interpretability for Neural Networks: Circuits, Sparse Features and Symbolic Reasoning](https://mlsanity.com/paper/2607.07316/) (2026)
- [Beyond Task Success: Behavioral and Representational Diagnostics for WAM and VLA](https://mlsanity.com/paper/2606.01095/) (2026)
- [Grounded Semantic Re-Binding for Robust Instruction Generalization in Vision-Language-Action Models](https://mlsanity.com/paper/2608.02497/) (2026)
- [IMPACT-VLA: Interaction-aware Multimodal Propagation Attribution via Counterfactual Trajectories for Vision-Language-Action Policies](https://mlsanity.com/paper/2609.15005/) (2026)
- [Natural State-Prediction Accuracy can Hide Weak Controlled Responsiveness in VLA Readouts](https://mlsanity.com/paper/2609.34684/) (2026)
- [eRLT: Efficient VLA Reinforcement Learning via Action-Relevant Token Routing](https://mlsanity.com/paper/2610.00913/) (2026)
- [Detect and Suppress: A Mechanistic Defense against Adversarial Patches in VLA Models](https://mlsanity.com/paper/2610.03498/) (2026)
- [Do VLAs Understand and Adapt to the Objects They Handle, or Simply Replay Learned Behaviors?](https://mlsanity.com/paper/2610.06078/) (2026)
- [When Listening Becomes Easier: Scrubbing Visual Cues for Shortcut-Free VLAs](https://mlsanity.com/paper/2610.10912/) (2026)

## Abstract (arXiv)

Vision-Language-Action (VLA) models have emerged as a promising approach for general-purpose robot manipulation. However, little research has mechanistically explored when and why they generalize across objects, scenes, and instructions. To probe internal representations, we train Sparse Autoencoders (SAEs) on the VLA's hidden-layer activations. SAEs learn sparse dictionaries over model activations, often revealing features that correspond to interpretable directions in the model's representation space. We identify SAE features corresponding to motion primitives and semantic concepts, including features that are general across episodes and causally steerable. We propose a metric to categorize features as general transferable primitives or episode-specific memorizations, offering a promising glimpse towards VLA generalization. We validate these findings through steering experiments on both the LIBERO simulation benchmark and on real-world DROID hardware. We find that amplifying general and semantic features induces behaviors consistent with their meanings, whereas ablating them destroys model performance. Furthermore, we demonstrate steering as a way to control behavior in unpromptable directions. Together, these results provide mechanistic evidence that VLAs can learn reusable internal features linking perception, language, and action across tasks and scenes. Our project page is located at https://drvla.github.io