When and How Severely: Scenario-Specific Safety Envelopes for Driving VLAs

Safety certification of Vision-Language-Action (VLA) driving planners under ISO 21448 (SOTIF) rests on an Operational Design Domain (ODD) specification that answers two complementary questions: when does the planner start to fail, and how severely does it fail once it does? We evaluate Alpamayo R1, a 10B-parameter open-weight driving VLA, on 15,968 (clip, attack) pairs. We find a conservative-aggregate gap: an aggregate safe threshold of $σ\leq 50$ under a 15% average displacement error (ADE) budget masks well-sampled scenarios that tolerate the top of the tested grid ($σ= 70$). A Gaussian Mixture Model (GMM) on the changed-explanation subset identifies six discrete severity bands (BIC-optimal $k{=}6$), so two perturbation conditions with the same mean error can differ materially in their share of high-severity (C4/C5) failures. Joining the two analyses on the same corpus surfaces a finding neither yields in isolation: the scenarios with the loosest noise thresholds are not those with the lowest high-severity rate: STOP_SIGNAL concentrates roughly $4\times$ the C4/C5 share of LANE_KEEPING despite tolerating a larger $σ$. A deployable SOTIF ODD specification for driving VLAs therefore requires a two-dimensional safety envelope, not a single aggregate value per hazard.

Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety

Safety standards for ML-based autonomous driving specify the kind of evidence an assurance case must contain (directed cause-and-effect chains, quantified interventional effects, named root-cause variables), yet the XAI literature is organised by output type and technique family (saliency maps, feature attribution, counterfactuals, causal graphs, language traces). SHAP, the most-recommended ADS XAI method, returns a ranked feature list that no implementation effort can convert into a directed chain (Fig.1). We name this mismatch the evidence-type gap. From AMLAS, ISO 26262, ISO21448, ISO/PAS 8800 we derive 19 testable evidentiary criteria across 7 lifecycle stages with representative clause-cited derivations and score six XAI method classes structurally. Causal XAI emerges as structurally required to satisfy the derived criteria at three stages: hazard identification (+62% rubric gap), incident investigation (+50%), and data management (+50%); the verdict set is stable across thresholds T in (0%, 50%]$ and survives a worst-case single-cell flip down to T = 25%. At the remaining four stages, correlational or language-based methods are comparable or sufficient. The rubric identifies structural admissibility (necessary but not sufficient for compliance): an admissible method's specific output content may still be wrong, and validating that fidelity (the edges a fitted SCM produces, the cause a trace names) is the open assurance challenge. A single-VLA proof of concept on 1,996 real-world driving clips (79,840 rows, ten splits) is consistent with each method's observed output type matching its rubric prediction. XAI method selection for ADS safety assurance should be driven by lifecycle-stage evidence demand, not by method popularity.

Lost in Fog: Sensor Perturbations Expose Reasoning Fragility in Driving VLAs

Interpretable autonomous driving planners depend not only on generating explanations, but also on those explanations remaining reliable under real-world sensor degradation. In this paper we present a controlled perturbation study of Vision-Language-Action (VLA) robustness in autonomous driving, evaluating Alpamayo R1 (10B parameters) across 1,996 scenarios under eight sensor perturbations (Gaussian noise at four intensities, two lighting extremes, and two fog levels; ${\sim}18{,}000$ inference trials). We find that reasoning consistency is a high-fidelity indicator of trajectory reliability: when Chain-of-Causation (CoC) explanations change after perturbation, trajectory deviation spikes $5.3{\times}$ (21.8m vs 4.1m), with $r\!=\!0.99$ across attack types and $r_{pb}\!=\!0.53$ per-sample (Cohen's $d\!=\!1.12$). A controlled ablation provides evidence that enabling CoC generation is associated with improved trajectory accuracy (11.8% on average across conditions; $p < 0.0001$) under matched inference settings. Over the tested noise range ($σ\in \{10, 30, 50, 70\}$), degradation is approximately linear ($R^2\!=\!0.957$), while standard input preprocessing defenses provide only marginal relief. Together, these results establish CoC consistency as a quantitative proxy for planning safety and motivate reasoning-based runtime monitoring for safer VLA deployment.

HySafe-AI: Hybrid Safety Architectural Analysis Framework for AI Systems: A Case Study

AI has become integral to safety-critical areas like autonomous driving systems (ADS) and robotics. The architecture of recent autonomous systems are trending toward end-to-end (E2E) monolithic architectures such as large language models (LLMs) and vision language models (VLMs). In this paper, we review different architectural solutions and then evaluate the efficacy of common safety analyses such as failure modes and effect analysis (FMEA) and fault tree analysis (FTA). We show how these techniques can be improved for the intricate nature of the foundational models, particularly in how they form and utilize latent representations. We introduce HySAFE-AI, Hybrid Safety Architectural Analysis Framework for AI Systems, a hybrid framework that adapts traditional methods to evaluate the safety of AI systems. Lastly, we offer hints of future work and suggestions to guide the evolution of future AI safety standards.

How much data do you need? Part 2: Predicting DL class specific training dataset sizes

This paper targets the question of predicting machine learning classification model performance, when taking into account the number of training examples per class and not just the overall number of training examples. This leads to the a combinatorial question, which combinations of number of training examples per class should be considered, given a fixed overall training dataset size. In order to solve this question, an algorithm is suggested which is motivated from special cases of space filling design of experiments. The resulting data are modeled using models like powerlaw curves and similar models, extended like generalized linear models i.e. by replacing the overall training dataset size by a parametrized linear combination of the number of training examples per label class. The proposed algorithm has been applied on the CIFAR10 and the EMNIST datasets.