Radu Calinescu

AAMAS Conference 2026 Conference Paper

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

• Daniel Bethell
• Simos Gerasimou
• Radu Calinescu
• Calum Imrie

Ensuring safe exploration in reinforcement learning is essential for real-world deployment. Existing methods, however, often trade safety for performance by producing overly conservative policies or diffuse cost estimates that weaken policy gradients. We propose the Uncertain Safety Critic (USC), which modulates conservatism usingcriticuncertaintyandrefinesunder-coveredregions, reducing safety violations by ≈ 40% while maintaining competitive or higher rewards and cutting cost-gradient error by ≈ 83%.

FM Conference 2026 Conference Paper

The SLEEC Framework for Normative Requirements Engineering

• Pedro Ribeiro
• Radu Calinescu
• Ana Cavalcanti
• Marsha Chechik
• Sinem Getir Yaman
• Lina Marsso
• Isobel Standen
• Beverley Townsend

Abstract Autonomous agents are increasingly deployed in sensitive, human-centric domains—such as healthcare, assistive care, and emergency response—where their decision-making must align with complex human norms. These translate into Social, Legal, Ethical, Empathetic, and Cultural (SLEEC) requirements that are often nuanced and context-dependent, challenging traditional software engineering paradigms. Our tutorial paper presents a comprehensive, tool-supported methodology for managing the SLEEC requirements lifecycle, covering elicitation, well-formedness validation, and conformance verification of software design models against SLEEC requirements. We demonstrate the use of our methodology and associated tools through application to a robot-assisted dressing system, providing a guide for researchers and engineers to bridge the gap between abstract human norms and verifiable system designs.

TAAS Journal 2025 Journal Article

An Architectural Viewpoint for Benefit-Cost-Risk-Aware Decision-Making in Self-Adaptive Systems

• Danny Weyns
• Sara M. Hezavehi
• Paris Avgeriou
• Radu Calinescu
• Raffaela Mirandola
• Diego Perez-Palacin

Self-adaptation equips a software system with a feedback loop that resolves uncertainties during operation and adapts the system to deal with them when necessary. Most self-adaptation approaches today use decision-making mechanisms that select for execution the adaptation option with the best-estimated benefit expressed as a set of adaptation goals. A few approaches also consider the estimated (one-off) cost of executing the candidate adaptation options. We argue that besides benefit and cost, decision-making in self-adaptive systems should also consider the estimated risk the system or its users would be exposed to if an adaptation option were selected for execution. Balancing all three concerns when evaluating the options for adaptation to mitigate uncertainty is essential for satisfying stakeholders’ concerns and ensuring the safety and public acceptance of self-adaptive systems. In this article, we present a reference model for decision-making in self-adaptation that considers the estimated benefit, cost, and risk as core concerns of each adaptation option. Leveraging this model, we then present an ISO/IEC/IEEE 42010 compatible architectural viewpoint that aims at supporting software architects responsible for designing robust decision-making mechanisms for self-adaptive systems. We demonstrate the applicability, usefulness, and understandability of the viewpoint through a case study where participants with experience in the engineering of self-adaptive systems performed a set of design tasks in DeltaIoT, an Internet-of-Things exemplar for research on self-adaptive systems.

TAAS Journal 2025 Journal Article

Efficient Model Verification at Runtime through Adaptive Dynamic Approximation

• Mehran Alidoost Nia
• Radu Calinescu
• Mehdi Kargahi
• Alessandro Abate

In dynamic environments, safety-critical autonomous systems must adapt to environmental changes without violating safety requirements. Model verification at runtime supports adaptation through the periodic analysis of continually updated models. A major limitation of the technique is the high overhead associated with the regular analyses of large state-space models. Our article introduces an adaptive approximation strategy that tackles this limitation by delaying unnecessary model updates, significantly reducing the overheads of these analyses. The strategy is applicable to Markov decision processes (MDPs) and is partitioned into components that can be analyzed independently and approximately. Each component is assigned a priority that depends on its impact on the accuracy of verification, and only the highest-priority components affected by changes are scheduled for updating/approximating. A complete update and verification of the entire model is triggered infrequently when the accuracy drops below a predefined threshold. We provide theoretical guarantees and proofs which ensure that our strategy can be applied without impacting the overall safety of the verified autonomous system. The experimental results from a case study in which we applied the strategy to a rescue robot team show that it is fully robust against safety-critical errors and can achieve a decision accuracy of over 97%.

TAAS Journal 2025 Journal Article

INSYTE: A Classification Framework for Traditional to Agentic AI Systems

• Zoe Porter
• Radu Calinescu
• Ernest Lim
• Victoria Hodge
• Philippa Ryan
• Simon Burton
• Ibrahim Habli
• Tom Lawton

Existing classification frameworks for AI and autonomous systems are being outpaced by recent advancements in AI technologies. This limits their applicability to modern intelligent systems, particularly agentic AI systems (autonomous systems that leverage foundation models to achieve wide-ranging, multi-layered goals). To address this deficiency, we introduce INSYTE, a multi-faceted framework that supports the classification of AI systems ranging from traditional rule-based systems to cutting-edge embodied AI and agentic systems. To that end, INSYTE considers the essential characteristics of an AI system across eight key dimensions grouped into four categories: system design ( underspecification and adaptiveness ); functionality ( breadth and depth ); operating environment ( diversity and dynamism ); and independence from human operational control ( intervention and oversight ). Different AI systems (or versions of systems) yield different ‘patterns’ on an eight-axis radar chart that INSYTE uses to provide an immediate visual summary of an AI system’s overall capability and a detailed representation of its individual characteristics. The INSYTE framework aligns with OECD’s definition of deployed AI systems, which is becoming the standard definition used by legislators and developers worldwide.

TAAS Journal 2025 Journal Article

Introduction to the SEAMS 2023 Special Issue

• Raffaela Mirandola
• Radu Calinescu
• Pooyan Jamshidi

TAAS Journal 2025 Journal Article

Parley +: Uncertainty Reduction in Self-Adaptive Systems

• Marc Carwehl
• Calum Imrie
• Thomas Vogel
• Genaína Rodrigues
• Radu Calinescu
• Lars Grunske

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 previous work proposed P arley, a more nuanced, adaptive approach to SAS uncertainty reduction. To that end, we introduced 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 with guarantees for the satisfaction of requirements. In this paper, we extend P arley to P arley + by improving the synthesis of these controllers, and by expanding the formalisation of P arley + to prove the validity of the synthesis. We illustrate the use and extend the evaluation of the effectiveness of our approach for mobile robot navigation and service-based system SAS. The evaluation results show that P arley + can synthesise controllers that help achieve the system’s objectives significantly better than P arley in 88.1% of the cases.

ECAI Conference 2025 Conference Paper

Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding

• Daniel Bethell
• Simos Gerasimou
• Radu Calinescu
• Calum Imrie

Safe exploration of reinforcement learning (RL) agents is a critical activity for empowering their deployment in many real-world scenarios. When prior knowledge of the target domain or task is unavailable, training RL agents in unknown, black-box environments unavoidably yields significant safety risks. Our ADVICE (Adaptive Shielding with a Contrastive Autoencoder) novel post-shielding approach operates in continuous state and action spaces, distinguishing safe and unsafe features of state-action pairs during training, and uses this knowledge to safeguard the RL agent from executing actions that yield likely hazardous outcomes. Our comprehensive experimental evaluation shows that ADVICE significantly reduces safety violations (≈50%) compared to state-of-the-art safe RL exploration approaches, while maintaining a competitive outcome reward for the synthesised safe policy.

TAAS Journal 2024 Journal Article

Predicting Nonfunctional Requirement Violations in Autonomous Systems

• Xinwei Fang
• Sinem Getir Yaman
• Radu Calinescu
• Julie Wilson
• Colin Paterson

Autonomous systems are often used in applications where environmental and internal changes may lead to requirement violations. Adapting to these changes proactively, i.e., before the violations occur, is preferable to recovering from the failures that may be caused by such violations. However, proactive adaptation needs methods for predicting requirement violations timely, accurately, and with acceptable overheads. To address this need, we present a method that allows autonomous systems to predict violations of performance, dependability and other nonfunctional requirements, and therefore take preventative measures to avoid or otherwise mitigate them. Our method for pre dicting these autonomou s sys t em disrupti o ns (PRESTO) comprises a design time stage and a run-time stage. At design-time, we use parametric model checking to obtain algebraic expressions that formalise the relationships between the nonfunctional properties of the requirements of interest (e.g., reliability, response time, and energy use) and the parameters of the system and its environment. At run-time, we predict future changes in these parameters by applying piece-wise linear regression to online data obtained through monitoring, and we use the algebraic expressions to predict the impact of these changes on the system requirements. We demonstrate the application of PRESTO through simulation in case studies from two different domains.

AAAI Conference 2024 Conference Paper

Robust Uncertainty Quantification Using Conformalised Monte Carlo Prediction

• Daniel Bethell
• Simos Gerasimou
• Radu Calinescu

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 comparable 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. The MC-CP code and replication package is available at https://github.com/team-daniel/MC-CP.

FormaliSE Conference 2023 Conference Paper

Goal Controller Synthesis for Self-Adaptive Systems

• Radu Calinescu
• Genaína Nunes Rodrigues

Much like humans, a growing number of software-controlled systems must cope with uncertainty and disruption through self-adaptation. However, while humans achieve this feat by changing both the means through which they pursue their goals and-when unavoidable-the goals themselves, self-adaptive systems are often only using the former adaptation mechanism. In this ‘research ideas’ paper, we argue that exploiting the latter mechanism is equally important, and we propose a new goal modelling paradigm that supports reasoning about goal change, and the use of probabilistic model checking tools to synthesise goal-management control software for self-adaptive systems.

TAAS Journal 2020 Journal Article

Uncertainty in Self-adaptive Systems: A Research Community Perspective

• Sara M. Hezavehi
• Danny Weyns
• Paris Avgeriou
• Radu Calinescu
• Raffaela Mirandola
• Diego Perez-Palacin

One of the primary drivers for self-adaptation is ensuring that systems achieve their goals regardless of the uncertainties they face during operation. Nevertheless, the concept of uncertainty in self-adaptive systems is still insufficiently understood. Several taxonomies of uncertainty have been proposed, and a substantial body of work exists on methods to tame uncertainty. Yet, these taxonomies and methods do not fully convey the research community’s perception on what constitutes uncertainty in self-adaptive systems and on the key characteristics of the approaches needed to tackle uncertainty. To understand this perception and learn from it, we conducted a survey comprising two complementary stages in which we collected the views of 54 and 51 participants, respectively. In the first stage, we focused on current research and development, exploring how the concept of uncertainty is understood in the community and how uncertainty is currently handled in the engineering of self-adaptive systems. In the second stage, we focused on directions for future research to identify potential approaches to dealing with unanticipated changes and other open challenges in handling uncertainty in self-adaptive systems. The key findings of the first stage are: (a) an overview of uncertainty sources considered in self-adaptive systems, (b) an overview of existing methods used to tackle uncertainty in concrete applications, (c) insights into the impact of uncertainty on non-functional requirements, (d) insights into different opinions in the perception of uncertainty within the community and the need for standardised uncertainty-handling processes to facilitate uncertainty management in self-adaptive systems. The key findings of the second stage are: (a) the insight that over 70% of the participants believe that self-adaptive systems can be engineered to cope with unanticipated change, (b) a set of potential approaches for dealing with unanticipated change, (c) a set of open challenges in mitigating uncertainty in self-adaptive systems, in particular in those with safety-critical requirements. From these findings, we outline an initial reference process to manage uncertainty in self-adaptive systems. We anticipate that the insights on uncertainty obtained from the community and our proposed reference process will inspire valuable future research on self-adaptive systems.