Aitor Arrieta

Arusa Kanwal, Pablo Valle, Shaukat Ali et al.

Vision-Language-Action (VLA) models are increasingly used as generalist robot policies, yet their evaluation still relies largely on static benchmarks that randomly sample task scenes. In high-dimensional embodied spaces, failures are sparse and clustered, so static benchmarking can underestimate robustness risks. We reframe VLA evaluation as an active failure-discovery problem and propose a failure-aware test-generation approach that combines diversity-driven exploration with surrogate models learned from observed executions. The method steers testing toward high-risk yet diverse scene regions. Across four state-of-the-art VLA models, it uncovers substantially more failures (up to +29.7 % over selected baselines) while revealing more diverse failure modes. This mean that, for instance, in the case of GR00T-N1.6, success rate dropped from 64.4% to 34.7%. More broadly, our findings call for a shift in VLA evaluation: from passive measurement on fixed task suites to adaptive, failure-seeking test generation that exposes the structure of model weaknesses before deployment.

Prasun Saurabh, Pablo Valle, Aitor Arrieta et al.

Testing robots requires assessing whether they perform their intended tasks correctly, dependably, and with high quality, a challenge known as the test oracle problem in software testing. Traditionally, this assessment relies on task-specific symbolic oracles for task correctness and on human manual evaluation of robot behavior, which is time-consuming, subjective, and error-prone. To address this, we propose VISOR, a Vision-Language Model (VLM)-based approach for automated test oracle assessment that eliminates the need of expensive human evaluations. VISOR performs automated evaluation of task correctness and quality, addressing the limitations of existing symbolic test oracles, which are task-specific and provide pass/fail judgments without explicitly quantifying task quality. Given the inherent uncertainty in VLMs, VISOR also explicitly quantifies its own uncertainty during test assessments. We evaluated VISOR using two VLMs, i.e., GPT and Gemini, across four robotic tasks on over 1,000 videos. Results show that Gemini achieves higher recall while GPT achieves higher precision. However, both models show low correlation between uncertainty and correctness, which prevents using uncertainty as a correctness predictor.

Muhammad Yousaf, Aitor Arrieta, Shaukat Ali et al.

Autonomous Underwater Robots (AURs) operate in challenging underwater environments, including low visibility and harsh water conditions. Such conditions present challenges for software engineers developing perception modules for the AUR software. To successfully carry out these tasks, deep learning has been incorporated into the AUR software to support its operations. However, the unique challenges of underwater environments pose difficulties for deep learning models, which often rely on labeled data that is scarce and noisy. This may undermine the trustworthiness of AUR software that relies on perception modules. Vision-Language Models (VLMs) offer promising solutions for AUR software as they generalize to unseen objects and remain robust in noisy conditions by inferring information from contextual cues. Despite this potential, their performance and uncertainty in underwater environments remain understudied from a software engineering perspective. Motivated by the needs of an industrial partner in assurance and risk management for maritime systems to assess the potential use of VLMs in this context, we present an empirical evaluation of VLM-based perception modules within the AUR software. We assess their ability to detect underwater trash by computing performance, uncertainty, and their relationship, to enable software engineers to select appropriate VLMs for their AUR software.

Aitor Arrieta, Miriam Ugarte, Pablo Valle et al.

The irruption of DeepSeek-R1 constitutes a turning point for the AI industry in general and the LLMs in particular. Its capabilities have demonstrated outstanding performance in several tasks, including creative thinking, code generation, maths and automated program repair, at apparently lower execution cost. However, LLMs must adhere to an important qualitative property, i.e., their alignment with safety and human values. A clear competitor of DeepSeek-R1 is its American counterpart, OpenAI's o3-mini model, which is expected to set high standards in terms of performance, safety and cost. In this technical report, we systematically assess the safety level of both DeepSeek-R1 (70b version) and OpenAI's o3-mini (beta version). To this end, we make use of our recently released automated safety testing tool, named ASTRAL. By leveraging this tool, we automatically and systematically generated and executed 1,260 test inputs on both models. After conducting a semi-automated assessment of the outcomes provided by both LLMs, the results indicate that DeepSeek-R1 produces significantly more unsafe responses (12%) than OpenAI's o3-mini (1.2%).

Pablo Valle, Shaukat Ali, Aitor Arrieta et al.

Vision-Language-Action (VLA) models are emerging robotic control systems that integrate perception, language understanding, and action generation in a unified architecture. Existing testing approaches for VLA-enabled robots rely on manually constructed symbolic test oracles that determine task success from final environment states. These oracles are costly to construct, require domain expertise, and are often tightly coupled to specific tasks and environments, limiting scalability and reuse. Furthermore, they provide only end-state assessments of task outcomes, offering limited insight into intermediate behavior and fault localization. To address these limitations, we introduce MANGO, a multi-agent framework that automatically generates fine-grained oracles from natural-language descriptions of robotic tasks. MANGO first generates a reusable library of atomic tasks, then generates simulator-grounded oracle definitions for each atomic task, and finally produces executable fine-grained oracles by decomposing complex instructions into ordered sequences of atomic actions and corresponding oracles. The framework uses collaborative Generator, Assessor, and Judge agents that iteratively refine generated artifacts through structured feedback. We evaluate MANGO on the LIBERO_10 and RoboCasa Humanoid Tabletop benchmarks. Results show that MANGO generates executable, fine-grained oracles that detect a similar number of failures as symbolic oracles while accurately localizing them and providing richer diagnostic information. Through ablation studies, we further analyzed component contributions and the effect of initial task set, while preserving oracle quality. Overall, the results show the feasibility and effectiveness of test oracle generation for VLA-enabled robots testing.

Aitor Arrieta, Miriam Ugarte, Pablo Valle et al.

Large Language Models (LLMs) have become an integral part of our daily lives. However, they impose certain risks, including those that can harm individuals' privacy, perpetuate biases and spread misinformation. These risks highlight the need for robust safety mechanisms, ethical guidelines, and thorough testing to ensure their responsible deployment. Safety of LLMs is a key property that needs to be thoroughly tested prior the model to be deployed and accessible to the general users. This paper reports the external safety testing experience conducted by researchers from Mondragon University and University of Seville on OpenAI's new o3-mini LLM as part of OpenAI's early access for safety testing program. In particular, we apply our tool, ASTRAL, to automatically and systematically generate up to date unsafe test inputs (i.e., prompts) that helps us test and assess different safety categories of LLMs. We automatically generate and execute a total of 10,080 unsafe test input on a early o3-mini beta version. After manually verifying the test cases classified as unsafe by ASTRAL, we identify a total of 87 actual instances of unsafe LLM behavior. We highlight key insights and findings uncovered during the pre-deployment external testing phase of OpenAI's latest LLM.

Miguel Romero-Arjona, Pablo Valle, Juan C. Alonso et al.

The battle for AI leadership is on, with OpenAI in the United States and DeepSeek in China as key contenders. In response to these global trends, the Spanish government has proposed ALIA, a public and transparent AI infrastructure incorporating small language models designed to support Spanish and co-official languages such as Basque. This paper presents the results of Red Teaming sessions, where ten participants applied their expertise and creativity to manually test three of the latest models from these initiatives$\unicode{x2013}$OpenAI o3-mini, DeepSeek R1, and ALIA Salamandra$\unicode{x2013}$focusing on biases and safety concerns. The results, based on 670 conversations, revealed vulnerabilities in all the models under test, with biased or unsafe responses ranging from 29.5% in o3-mini to 50.6% in Salamandra. These findings underscore the persistent challenges in developing reliable and trustworthy AI systems, particularly those intended to support Spanish and Basque languages.

Miriam Ugarte, Pablo Valle, José Antonio Parejo et al.

Large Language Models (LLMs) have recently gained attention due to their ability to understand and generate sophisticated human-like content. However, ensuring their safety is paramount as they might provide harmful and unsafe responses. Existing LLM testing frameworks address various safety-related concerns (e.g., drugs, terrorism, animal abuse) but often face challenges due to unbalanced and obsolete datasets. In this paper, we present ASTRAL, a tool that automates the generation and execution of test cases (i.e., prompts) for testing the safety of LLMs. First, we introduce a novel black-box coverage criterion to generate balanced and diverse unsafe test inputs across a diverse set of safety categories as well as linguistic writing characteristics (i.e., different style and persuasive writing techniques). Second, we propose an LLM-based approach that leverages Retrieval Augmented Generation (RAG), few-shot prompting strategies and web browsing to generate up-to-date test inputs. Lastly, similar to current LLM test automation techniques, we leverage LLMs as test oracles to distinguish between safe and unsafe test outputs, allowing a fully automated testing approach. We conduct an extensive evaluation on well-known LLMs, revealing the following key findings: i) GPT3.5 outperforms other LLMs when acting as the test oracle, accurately detecting unsafe responses, and even surpassing more recent LLMs (e.g., GPT-4), as well as LLMs that are specifically tailored to detect unsafe LLM outputs (e.g., LlamaGuard); ii) the results confirm that our approach can uncover nearly twice as many unsafe LLM behaviors with the same number of test inputs compared to currently used static datasets; and iii) our black-box coverage criterion combined with web browsing can effectively guide the LLM on generating up-to-date unsafe test inputs, significantly increasing the number of unsafe LLM behaviors.

Jiahui Wu, Chengjie Lu, Aitor Arrieta et al.

Autonomous vehicles (AVs) make driving decisions without human intervention. Therefore, ensuring AVs' dependability is critical. Despite significant research and development in AV development, their dependability assurance remains a significant challenge due to the complexity and unpredictability of their operating environments. Scenario-based testing evaluates AVs under various driving scenarios, but the unlimited number of potential scenarios highlights the importance of identifying critical scenarios that can violate safety or functional requirements. Such requirements are inherently interdependent and need to be tested simultaneously. To this end, we propose MOEQT, a novel multi-objective reinforcement learning (MORL)-based approach to generate critical scenarios that simultaneously test interdependent safety and functional requirements. MOEQT adapts Envelope Q-learning as the MORL algorithm, which dynamically adapts multi-objective weights to balance the relative importance between multiple objectives. MOEQT generates critical scenarios to violate multiple requirements through dynamically interacting with the AV environment, ensuring comprehensive AV testing. We evaluate MOEQT using an advanced end-to-end AV controller and a high-fidelity simulator and compare MOEQT with two baselines: a random strategy and a single-objective RL with a weighted reward function. Our evaluation results show that MOEQT achieved an overall better performance in identifying critical scenarios for violating multiple requirements than the baselines.

Aitor Arrieta

Cyber-Physical Systems (CPSs) combine software and physical components. These systems are widely applied in society within many domains, including the automotive, aerospace, railway, etc. Testing these systems is extremely challenging, therefore, it has attracted significant attention from the research community. A driving CPS testing technique in industry is simulation-based testing. However, this poses significant challenges. In this new-idea paper we present a novel approach to enhance the testing processes of CPSs. This novel approach is motivated with examples and open questions.