Detecting and Explaining Robot Policy Failures from Sparse Internal Features
cs.RO
Submitted: 2026-11-07
Updated: 2026-11-07
Terminology
Sources
- TAIL-Safe: Task-Agnostic Safety Monitoring for Imitation Learning Policies
- Understanding intermediate layers using linear classifier probes
- $\pi_0$: A Vision-Language-Action Flow Model for General Robot Control
- RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control
- Observing and Controlling Features in Vision-Language-Action Models
- RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulation
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- LIBERO-Plus: In-depth Robustness Analysis of Vision-Language-Action Models
- Scaling and evaluating sparse autoencoders
- Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models
- Event-Grounded Sparse Autoencoders for Vision-Language-Action Policies
- Probing a Vision-Language-Action Model for Symbolic States and Integration into a Cognitive Architecture
- Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models
- Emergent World Representations in OpenVLA
- $\pi_{0.5}$: a Vision-Language-Action Model with Open-World Generalization
- SAM 2: Segment Anything in Images and Videos
- VLA-FAIL: Efficient Task Failure Detection for Finetuned Vision-Language-Action Models
- Sparse Autoencoders Reveal Interpretable and Steerable Features in VLA Models
- Steering Language Models With Activation Engineering
- A Pragmatic VLA Foundation Model
Related papers
- Rethinking Actions: Continuity-Aware Priors for World-Model Planning
- MOSAIC: Modular Obstacle and Spatial-Aware Framework for Manipulation In Cluttered Scenes
- SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
- DexSeed: Expanding Human Hand-Object Interactions into Robot Demonstrations for Dexterous Policy Learning
- LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
- CIALLO: Learning Shared Intent Representations for Cross-Embodiment Manipulation