REVIEW 14 cited by
Can We Detect Failures Without Failure Data? Uncertainty-Aware Runtime Failure Detection for Imitation Learning Policies
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
Can We Detect Failures Without Failure Data? Uncertainty-Aware Runtime Failure Detection for Imitation Learning Policies
read the original abstract
Recent years have witnessed impressive robotic manipulation systems driven by advances in imitation learning and generative modeling, such as diffusion- and flow-based approaches. As robot policy performance increases, so does the complexity and time horizon of achievable tasks, inducing unexpected and diverse failure modes that are difficult to predict a priori. To enable trustworthy policy deployment in safety-critical human environments, reliable runtime failure detection becomes important during policy inference. However, most existing failure detection approaches rely on prior knowledge of failure modes and require failure data during training, which imposes a significant challenge in practicality and scalability. In response to these limitations, we present FAIL-Detect, a modular two-stage approach for failure detection in imitation learning-based robotic manipulation. To accurately identify failures from successful training data alone, we frame the problem as sequential out-of-distribution (OOD) detection. We first distill policy inputs and outputs into scalar signals that correlate with policy failures and capture epistemic uncertainty. FAIL-Detect then employs conformal prediction (CP) as a versatile framework for uncertainty quantification with statistical guarantees. Empirically, we thoroughly investigate both learned and post-hoc scalar signal candidates on diverse robotic manipulation tasks. Our experiments show learned signals to be mostly consistently effective, particularly when using our novel flow-based density estimator. Furthermore, our method detects failures more accurately and faster than state-of-the-art (SOTA) failure detection baselines. These results highlight the potential of FAIL-Detect to enhance the safety and reliability of imitation learning-based robotic systems as they progress toward real-world deployment.
Forward citations
Cited by 14 Pith papers
-
ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policies
ActProbe is an action-space detector that uses temporal consistency error and action chunk magnitude from policy outputs, mapped via LSTM-MLP, to predict failures earlier than baselines across policies and real-robot tasks.
-
Failure Identification in Imitation Learning Via Statistical and Semantic Filtering
FIDeL detects failures in imitation learning by building compact nominal representations via optimal transport, applying conformal prediction thresholds, and using VLMs for semantic filtering, outperforming baselines ...
-
The Geometric Nature and a Free Proxy for Flow-Matching Uncertainty
A single-pass measure of how much a flow-matching action trajectory bends correlates with the model's uncertainty and can flag impending robot failures for free.
-
RL$^2$-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action Models
Adaptive compositional steering of frozen VLAs with a latent-conditioned offline RL flow policy, gated by failure prediction, improves OOD manipulation success by up to +17.3%.
-
Foresight: Failure Detection for Long-Horizon Robotic Manipulation with Action-Conditioned World Model Latents
Foresight detects failures in long-horizon robotic manipulation using latents from action-conditioned world models trained only on task-level labels and calibrated via functional conformal prediction.
-
Robot Critics that Sweat the Small Stuff
Fine-tuning VLMs with pairwise progress supervision from policy rollouts improves fine-grained failure detection and boosts robot manipulation success by 11% real-world and 5.9% in simulation.
-
Tri-Info: Generalizable, Interpretable Failure Prediction for VLA Models via Information Theory
Tri-Info uses three information theory signals on action diversity, temporal consistency, and state coupling to predict VLA model failures with cross-domain generalization to 83% real-world accuracy.
-
AEGIS: A Backup Reflex for Physical AI
AEGIS uses activation probes for early-warning detection of high-risk steps in weak policies and selectively escalates to stronger policies, recovering 10.1% of lost trajectories on LIBERO-Spatial while activating the...
-
Flow-based Policy Adaptation without Policy Updates
GLOVES learns flow models from limited expert demonstrations to selectively correct actions from non-expert policies or operators toward expert distributions using reverse-flow OOD detection as an intervention gate.
-
VLAConf: Calibrated Task-Success Confidence for Vision-Language-Action Models
VLAConf is a one-class discriminative method that estimates step-wise task-success confidence for VLA models via anomaly scoring on frozen representations plus step-conditioned modeling, shown to be more efficient tha...
-
CycleVLA: Proactive Self-Correcting Vision-Language-Action Models via Subtask Backtracking and Minimum Bayes Risk Decoding
CycleVLA adds progress-triggered VLM failure checks, subtask backtracking, and MBR consensus decoding to VLAs, raising LIBERO average success from 89.3% to 95.3% and claiming 91% real-robot success.
-
InFeR: Informed Failure Resilience in Learned Visual Navigation Control
InFeR retrains imitation learning policies with a VIB loss for OOD failure detection and applies Grad-CAM to localize failure sources, enabling heuristic recovery in visual navigation without additional demonstrations.
-
INSIGHT: INference-time Sequence Introspection for Generating Help Triggers in Vision-Language-Action Models
Token-level uncertainty sequences from a VLA policy, classified by a small transformer, predict when a robot should request human help better than static uncertainty scores.
-
RL$^2$-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action Models
RL² improves VLA robot success rates by conditionally composing an offline RL policy's actions with the frozen VLA only when a failure detector flags impending failure.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.