arXiv subjects

Radu Calinescu

Publications and source records attributed to Radu Calinescu.

At least 19 recordsLinked to original sources

arXiv2026

Mind the Prompt: Self-adaptive Generation of Task Plan Explanations via LLMs

Integrating Large Language Models (LLMs) into complex software systems enables the generation of human-understandable explanations of opaque AI processes, such as automated task planning. However, the quality and reliability of these explanations heavily depend on effective prompt engineering. The lack of a systematic understanding of how diverse stakeholder groups formulate and refine prompts hinders the development of tools that can automate this process. We introduce COMPASS (COgnitive Modelling for Prompt Automated SynthesiS), a proof-of-concept self-adaptive approach that formalises prompt engineering as a cognitive and probabilistic decision-making process. COMPASS models unobservable users' latent cognitive states, such as attention and comprehension, uncertainty, and observable interaction cues as a POMDP, whose synthesised policy enables adaptive generation of explanations and prompt refinements. We evaluate COMPASS using two diverse cyber-physical system case studies to assess the adaptive explanation generation and their qualities, both quantitatively and qualitatively. Our results demonstrate the feasibility of COMPASS integrating human cognition and user profile's feedback into automated prompt synthesis in complex task planning systems.

cs.AI↗

arXiv2026

Hazard Management in Robot-Assisted Mammography Support

Robotic and embodied-AI systems have the potential to improve accessibility and quality of care in clinical settings, but their deployment in close physical contact with vulnerable patients introduces significant safety risks. This paper presents a hazard management methodology for MammoBot, an assistive robotic system designed to support patients during X-ray mammography. To ensure safety from early development stages, we combine stakeholder-guided process modelling with Software Hazard Analysis and Resolution in Design (SHARD) and System-Theoretic Process Analysis (STPA). The robot-assisted workflow is defined collaboratively with clinicians, roboticists, and patient representatives to capture key human-robot interactions. SHARD is applied to identify technical and procedural deviations, while STPA is used to analyse unsafe control actions arising from user interaction. The results show that many hazards arise not from component failures, but from timing mismatches, premature actions, and misinterpretation of system state. These hazards are translated into refined and additional safety requirements that constrain system behaviour and reduce reliance on correct human timing or interpretation alone. The work demonstrates a structured and traceable approach to safety-driven design with potential applicability to assistive robotic systems in clinical environments.

cs.RO↗

arXiv2026

Social, Legal, Ethical, Empathetic and Cultural Norm Operationalisation for AI Agents

As AI agents are increasingly used in high-stakes domains like healthcare and law enforcement, aligning their behaviour with social, legal, ethical, empathetic, and cultural (SLEEC) norms has become a critical engineering challenge. While international frameworks have established high-level normative principles for AI, a significant gap remains in translating these abstract principles into concrete, verifiable requirements. To address this gap, we propose a systematic SLEEC-norm operationalisation process for determining, validating, implementing, and verifying normative requirements. Furthermore, we survey the landscape of methods and tools supporting this process, and identify key remaining challenges and research avenues for addressing them. We thus establish a framework - and define a research and policy agenda - for developing AI agents that are not only functionally useful but also demonstrably aligned with human norms and values.

cs.AI↗

arXiv2026

Interpretable Attention-Based Multi-Agent PPO for Latency Spike Resolution in 6G RAN Slicing

Sixth-generation (6G) radio access networks (RANs) must enforce strict service-level agreements (SLAs) for heterogeneous slices, yet sudden latency spikes remain difficult to diagnose and resolve with conventional deep reinforcement learning (DRL) or explainable RL (XRL). We propose \emph{Attention-Enhanced Multi-Agent Proximal Policy Optimization (AE-MAPPO)}, which integrates six specialized attention mechanisms into multi-agent slice control and surfaces them as zero-cost, faithful explanations. The framework operates across O-RAN timescales with a three-phase strategy: predictive, reactive, and inter-slice optimization. A URLLC case study shows AE-MAPPO resolves a latency spike in $18$ms, restores latency to $0.98$ms with $99.9999\%$ reliability, and reduces troubleshooting time by $93\%$ while maintaining eMBB and mMTC continuity. These results confirm AE-MAPPO's ability to combine SLA compliance with inherent interpretability, enabling trustworthy and real-time automation for 6G RAN slicing.

eess.SY↗

arXiv2025

Automated Formalization of Probabilistic Requirements from Structured Natural Language

Integrating autonomous and adaptive behavior into software-intensive systems presents significant challenges for software development, as uncertainties in the environment or decision-making processes must be explicitly captured. These challenges are amplified in safety- and mission-critical systems, which must undergo rigorous scrutiny during design and development. Key among these challenges is the difficulty of specifying requirements that use probabilistic constructs to capture the uncertainty affecting these systems. To enable formal analysis, such requirements must be expressed in precise mathematical notations such as probabilistic logics. However, expecting developers to write requirements directly in complex formalisms is unrealistic and highly error-prone. We extend the structured natural language used by NASA's Formal Requirement Elicitation Tool (FRET) with support for the specification of unambiguous and correct probabilistic requirements, and develop an automated approach for translating these requirements into logical formulas. We propose and develop a formal, compositional, and automated approach for translating structured natural-language requirements into formulas in probabilistic temporal logic. To increase trust in our formalizations, we provide assurance that the generated formulas are well-formed and conform to the intended semantics through an automated validation framework and a formal proof. The extended FRET tool enables developers to specify probabilistic requirements in structured natural language, and to automatically translate them into probabilistic temporal logic, making the formal analysis of autonomous and adaptive systems more practical and less error-prone.

cs.PL↗

arXiv2025

Safe But Not Sorry: Reducing Over-Conservatism in Safety Critics via Uncertainty-Aware Modulation

Ensuring the safe exploration of reinforcement learning (RL) agents is critical for deployment in real-world systems. Yet existing approaches struggle to strike the right balance: methods that tightly enforce safety often cripple task performance, while those that prioritize reward leave safety constraints frequently violated, producing diffuse cost landscapes that flatten gradients and stall policy improvement. We introduce the Uncertain Safety Critic (USC), a novel approach that integrates uncertainty-aware modulation and refinement into critic training. By concentrating conservatism in uncertain and costly regions while preserving sharp gradients in safe areas, USC enables policies to achieve effective reward-safety trade-offs. Extensive experiments show that USC reduces safety violations by approximately 40% while maintaining competitive or higher rewards, and reduces the error between predicted and true cost gradients by approximately 83%, breaking the prevailing trade-off between safety and performance and paving the way for scalable safe RL.

cs.LG↗

arXiv2025

Learning to Navigate Under Imperfect Perception: Conformalised Segmentation for Safe Reinforcement Learning

Reliable navigation in safety-critical environments requires both accurate hazard perception and principled uncertainty handling to strengthen downstream safety handling. Despite the effectiveness of existing approaches, they assume perfect hazard detection capabilities, while uncertainty-aware perception approaches lack finite-sample guarantees. We present COPPOL, a conformal-driven perception-to-policy learning approach that integrates distribution-free, finite-sample safety guarantees into semantic segmentation, yielding calibrated hazard maps with rigorous bounds for missed detections. These maps induce risk-aware cost fields for downstream RL planning. Across two satellite-derived benchmarks, COPPOL increases hazard coverage (up to 6x) compared to comparative baselines, achieving near-complete detection of unsafe regions while reducing hazardous violations during navigation (up to approx 50%). More importantly, our approach remains robust to distributional shift, preserving both safety and efficiency.

cs.LG↗

arXiv2025

Tool for Supporting Debugging and Understanding of Normative Requirements Using LLMs

Normative requirements specify social, legal, ethical, empathetic, and cultural (SLEEC) norms that must be observed by a system. To support the identification of SLEEC requirements, numerous standards and regulations have been developed. These requirements are typically defined by stakeholders in the non-technical system with diverse expertise (e.g., ethicists, lawyers, social scientists). Hence, ensuring their consistency and managing the requirement elicitation process are complex and error-prone tasks. Recent research has addressed this challenge using domain-specific languages to specify normative requirements as rules, whose consistency can then be analyzed with formal methods. Nevertheless, these approaches often present the results from formal verification tools in a way that is inaccessible to non-technical users. This hinders understanding and makes the iterative process of eliciting and validating these requirements inefficient in terms of both time and effort. To address this problem, we introduce SLEEC-LLM, a tool that uses large language models (LLMs) to provide natural-language interpretations for model-checking counterexamples corresponding to SLEEC rule inconsistencies. SLEEC-LLM improves the efficiency and explainability of normative requirements elicitation and consistency analysis. To demonstrate its effectiveness, we summarise its use in two real-world case studies involving non-technical stakeholders.

cs.SE↗

arXiv2025

Conformal Safety Shielding for Imperfect-Perception Agents

We consider the problem of safe control in discrete autonomous agents that use learned components for imperfect perception (or more generally, state estimation) from high-dimensional observations. We propose a shield construction that provides run-time safety guarantees under perception errors by restricting the actions available to an agent, modeled as a Markov decision process, as a function of the state estimates. Our construction uses conformal prediction for the perception component, which guarantees that for each observation, the predicted set of estimates includes the actual state with a user-specified probability. The shield allows an action only if it is allowed for all the estimates in the predicted set, resulting in local safety. We also articulate and prove a global safety property of existing shield constructions for perfect-perception agents bounding the probability of reaching unsafe states if the agent always chooses actions prescribed by the shield. We illustrate our approach with a case-study of an experimental autonomous system that guides airplanes on taxiways using high-dimensional perception DNNs.

eess.SY↗

arXiv2025

Symbolic Runtime Verification and Adaptive Decision-Making for Robot-Assisted Dressing

We present a control framework for robot-assisted dressing that augments low-level hazard response with runtime monitoring and formal verification. A parametric discrete-time Markov chain (pDTMC) models the dressing process, while Bayesian inference dynamically updates this pDTMC's transition probabilities based on sensory and user feedback. Safety constraints from hazard analysis are expressed in probabilistic computation tree logic, and symbolically verified using a probabilistic model checker. We evaluate reachability, cost, and reward trade-offs for garment-snag mitigation and escalation, enabling real-time adaptation. Our approach provides a formal yet lightweight foundation for safety-aware, explainable robotic assistance.

cs.RO↗

arXiv2025

Verification and External Parameter Inference for Stochastic World Models

Given its ability to analyse stochastic models ranging from discrete and continuous-time Markov chains to Markov decision processes and stochastic games, probabilistic model checking (PMC) is widely used to verify system dependability and performance properties. However, modelling the behaviour of, and verifying these properties for many software-intensive systems requires the joint analysis of multiple interdependent stochastic models of different types, which existing PMC techniques and tools cannot handle. To address this limitation, we introduce a tool-supported UniversaL stochasTIc Modelling, verificAtion and synThEsis (ULTIMATE) framework that supports the representation, verification and synthesis of heterogeneous multi-model stochastic systems with complex model interdependencies. Through its unique integration of multiple PMC paradigms, and underpinned by a novel verification method for handling model interdependencies, ULTIMATE unifies-for the first time-the modelling of probabilistic and nondeterministic uncertainty, discrete and continuous time, partial observability, and the use of both Bayesian and frequentist inference to exploit domain knowledge and data about the modelled system and its context. A comprehensive suite of case studies and experiments confirm the generality and effectiveness of our novel verification framework.

cs.LO↗

arXiv2024

Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding

Empowering safe exploration of reinforcement learning (RL) agents during training is a critical challenge towards their deployment in many real-world scenarios. When prior knowledge of the domain or task is unavailable, training RL agents in unknown, black-box environments presents an even greater safety risk. We introduce ADVICE (Adaptive Shielding with a Contrastive Autoencoder), a novel post-shielding technique that distinguishes safe and unsafe features of state-action pairs during training, and uses this knowledge to protect the RL agent from executing actions that yield likely hazardous outcomes. Our comprehensive experimental evaluation against state-of-the-art safe RL exploration techniques shows that ADVICE significantly reduces safety violations (approx 50%) during training, with a competitive outcome reward compared to other techniques.

cs.AI↗

arXiv2024

Normative Requirements Operationalization with Large Language Models

Normative non-functional requirements specify constraints that a system must observe in order to avoid violations of social, legal, ethical, empathetic, and cultural norms. As these requirements are typically defined by non-technical system stakeholders with different expertise and priorities (ethicists, lawyers, social scientists, etc.), ensuring their well-formedness and consistency is very challenging. Recent research has tackled this challenge using a domain-specific language to specify normative requirements as rules whose consistency can then be analysed with formal methods. In this paper, we propose a complementary approach that uses Large Language Models to extract semantic relationships between abstract representations of system capabilities. These relations, which are often assumed implicitly by non-technical stakeholders (e.g., based on common sense or domain knowledge), are then used to enrich the automated reasoning techniques for eliciting and analyzing the consistency of normative requirements. We show the effectiveness of our approach to normative requirements elicitation and operationalization through a range of real-world case studies.

cs.SE↗

arXiv2024

Formal Synthesis of Uncertainty Reduction Controllers

In its quest for approaches to taming uncertainty in self-adaptive systems (SAS), the research community has largely focused on solutions that adapt the SAS architecture or behaviour in response to uncertainty. By comparison, solutions that reduce the uncertainty affecting SAS (other than through the blanket monitoring of their components and environment) remain underexplored. Our paper proposes a more nuanced, adaptive approach to SAS uncertainty reduction. To that end, we introduce a SAS architecture comprising an uncertainty reduction controller that drives the adaptive acquisition of new information within the SAS adaptation loop, and a tool-supported method that uses probabilistic model checking to synthesise such controllers. The controllers generated by our method deliver optimal trade-offs between SAS uncertainty reduction benefits and new information acquisition costs. We illustrate the use and evaluate the effectiveness of our approach for mobile robot navigation and server infrastructure management SAS.

cs.SE↗

arXiv2024

Analyzing and Debugging Normative Requirements via Satisfiability Checking

As software systems increasingly interact with humans in application domains such as transportation and healthcare, they raise concerns related to the social, legal, ethical, empathetic, and cultural (SLEEC) norms and values of their stakeholders. Normative non-functional requirements (N-NFRs) are used to capture these concerns by setting SLEEC-relevant boundaries for system behavior. Since N-NFRs need to be specified by multiple stakeholders with widely different, non-technical expertise (ethicists, lawyers, regulators, end users, etc.), N-NFR elicitation is very challenging. To address this challenge, we introduce N-Check, a novel tool-supported formal approach to N-NFR analysis and debugging. N-Check employs satisfiability checking to identify a broad spectrum of N-NFR well-formedness issues (WFI), such as conflicts, redundancy, restrictiveness, insufficiency, yielding diagnostics which pinpoint their causes in a user-friendly way that enables non-technical stakeholders to understand and fix them. We show the effectiveness and usability of our approach through nine case studies in which teams of ethicists, lawyers, philosophers, psychologists, safety analysts, and engineers used N-Check to analyse and debug 233 N-NFRs comprising 62 issues for the software underpinning the operation of systems ranging from assistive-care robots and tree-disease detection drones to manufacturing collaborative robots.

cs.SE↗

arXiv2023

Out-of-distribution Object Detection through Bayesian Uncertainty Estimation

The superior performance of object detectors is often established under the condition that the test samples are in the same distribution as the training data. However, in many practical applications, out-of-distribution (OOD) instances are inevitable and usually lead to uncertainty in the results. In this paper, we propose a novel, intuitive, and scalable probabilistic object detection method for OOD detection. Unlike other uncertainty-modeling methods that either require huge computational costs to infer the weight distributions or rely on model training through synthetic outlier data, our method is able to distinguish between in-distribution (ID) data and OOD data via weight parameter sampling from proposed Gaussian distributions based on pre-trained networks. We demonstrate that our Bayesian object detector can achieve satisfactory OOD identification performance by reducing the FPR95 score by up to 8.19% and increasing the AUROC score by up to 13.94% when trained on BDD100k and VOC datasets as the ID datasets and evaluated on COCO2017 dataset as the OOD dataset.

cs.CV↗

arXiv2023

Robust Uncertainty Quantification Using Conformalised Monte Carlo Prediction

Deploying deep learning models in safety-critical applications remains a very challenging task, mandating the provision of assurances for the dependable operation of these models. Uncertainty quantification (UQ) methods estimate the model's confidence per prediction, informing decision-making by considering the effect of randomness and model misspecification. Despite the advances of state-of-the-art UQ methods, they are computationally expensive or produce conservative prediction sets/intervals. We introduce MC-CP, a novel hybrid UQ method that combines a new adaptive Monte Carlo (MC) dropout method with conformal prediction (CP). MC-CP adaptively modulates the traditional MC dropout at runtime to save memory and computation resources, enabling predictions to be consumed by CP, yielding robust prediction sets/intervals. Throughout comprehensive experiments, we show that MC-CP delivers significant improvements over advanced UQ methods, like MC dropout, RAPS and CQR, both in classification and regression benchmarks. MC-CP can be easily added to existing models, making its deployment simple.

cs.LG↗

arXiv2023

Specification, Validation and Verification of Social, Legal, Ethical, Empathetic and Cultural Requirements for Autonomous Agents

Autonomous agents are increasingly being proposed for use in healthcare, assistive care, education, and other applications governed by complex human-centric norms. To ensure compliance with these norms, the rules they induce need to be unambiguously defined, checked for consistency, and used to verify the agent. In this paper, we introduce a framework for formal specification, validation and verification of social, legal, ethical, empathetic and cultural (SLEEC) rules for autonomous agents. Our framework comprises: (i) a language for specifying SLEEC rules and rule defeaters (that is, circumstances in which a rule does not apply or an alternative form of the rule is required); (ii) a formal semantics (defined in the process algebra tock-CSP) for the language; and (iii) methods for detecting conflicts and redundancy within a set of rules, and for verifying the compliance of an autonomous agent with such rules. We show the applicability of our framework for two autonomous agents from different domains: a firefighter UAV, and an assistive-dressing robot.

cs.SE↗