REVIEW 6 cited by
Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models
read the original abstract
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
Forward citations
Cited by 6 Pith papers
-
Point Tracking Improves World Action Models
JOPAT jointly models pixels, point tracks, and actions in a diffusion transformer and reports gains over pixel-only baselines on long-horizon robot tasks with occlusion and off-screen motion.
-
Event-Grounded Sparse Autoencoders for Vision-Language-Action Policies
Event-grounded SAE analysis in VLA policies produces stronger causal effects on robot behavior than standard methods by anchoring features to clustered end-effector keyframes across simulations and real-robot tests.
-
Sequential Planning via Anchored Robotic Keypoints
SPARK reaches 43.7% success on six LIBERO-PRO cells by LLM-generated typed behavior trees plus multi-prompt perception and recovery, more than doubling CaP-Agent0 and VLA baselines.
-
Learning What to Say to Your VLA: Mostly Harmless Vision Language Action Model Steering
A search-and-distill framework with conformalized improvement head produces a language feedback policy that boosts frozen VLA performance by 24.7% in simulation and 65% on hardware while guaranteeing harmlessness on p...
-
Beyond Task Success: Behavioral and Representational Diagnostics for WAM and VLA
Empirical study introduces behavioral and representational diagnostics showing architecture-dependent gains in object targeting and predictive structure for WAMs over VLAs on LIBERO and RoboTwin2.0.
-
VLA-Trace: Diagnosing Vision-Language-Action Models through Representation and Behavior Tracing
VLA-Trace diagnoses two VLA models via representation CKA, attention interventions, and behavioral tests, finding distinct finetuning dynamics, different routing, and strong visual grounding but weak fine-grained sema...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.