Distributional Uncertainty and Adaptive Decision-Making in System Co-design

Yujun Huang, Gioele Zardini · 2026

Read Distributional Uncertainty and Adaptive Decision-Making in System Co-design by Yujun Huang, Gioele Zardini (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-14047.txt.

Introduction

Huang and Zardini extend monotone co-design so engineered systems can be planned under chance-based uncertainty. Older co-design models use interval bounds, which support worst-case thinking but not risk or staged choices. Their version treats each design outcome as a distribution over design problems. Standard co-design combinations still compose cleanly in this richer setup. New queries report feasibility chances, confidence bounds, and resource needs. A drone case study shows risk-aware choices that interval models cannot state.

Catalog record

CREATORYujun Huang, Gioele Zardini
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE120418
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.14047

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Distributional Uncertainty and Adaptive Decision-Making in System Co-design Yujun Huang, Gioele Zardini Abstract Complex engineered systems require coordinated design choices across heterogeneous components under multiple conflicting objectives and uncertain specifications. Monotone co-design provides a compositional framework for such problems by modeling each subsystem as a design problem: a feasible relation between provided functionalities and required resources in partially ordered sets. Existing uncertain co-design models rely on interval bounds, which support worst-case reasoning but cannot represent probabilistic risk or multi-stage adaptive decisions. We develop a distributional extension of co-design that models uncertain design outcomes as distributions over design problems and supports adaptive decision processes through Markov-kernel re-parameterizations. Using quasi-measurable and quasi-universal spaces, we show that the standard co-design interconnection operations remain compositional under this richer notion of uncertainty. We further introduce queries and observations that extract probabilistic design trade-offs, including feasibility probabilities, confidence bounds, and distributions of minimal required resources. A task-driven unmanned aerial vehicle case study illustrates how the framework captures risk-sensitive and information-dependent design choices that interval-based models cannot express. I Introduction Embodied intelligence and cyber-physical systems (CPS) tightly couple heterogeneous hardware (sensors, actuators, processors) with software (perception, planning, control), and must be evaluated against multiple, often incomparable, metrics. As a result, designers face nontrivial trade-offs across components, subsystems, and system-level objectives, while coordinating stakeholders with diverse expertise. For safety- and mission-critical applications, uncertainty cannot be treated purely adversarially: designers must quantify risk (e.g., probability of meeting specifications) and reason about how robustness trades against performance. Moreover, design is rarely one-shot: decisions are made in stages as information becomes available, so later choices should be adaptive to earlier commitments and intermediate observations. Monotone co-design provides a compositional framework for such multiobjective system design [ 7 , 28 , 6 ] . Each component is modeled as a design problem (DP): a feasible relation between provided functionalities and required resources, ordered in partially ordered sets. Systems are constructed by interconnecting DPsin series, parallel, and feedback, supported by traced and symmetric monoidal categorical structure [ 28 ] . This common language has enabled joint optimization of heterogeneous hardware/software architectures across robotics and control [ 25 , 27 , 20 ] , transportation [ 26 ] , and automotive [ 21 ] . Existing co-design tools, however, treat uncertainty primarily via interval bounds on feasible sets [ 8 , 28 ] . Intervals yield worst-case robustness but are too coarse to encode probabilistic risk (chance constraints, confidence levels, quantiles) and do not natively express multi-stage adaptive decision policies. In particular, they cannot distinguish “almost-always safe” designs from uniformly conservative ones, nor represent strategies that react to intermediate observations (e.g., test outcomes, demand realizations, partial measurements). This paper extends monotone co-design to a distributional and adaptive uncertainty framework. We model uncertain specifications and outcomes via probability measures (and parametric families) over spaces of DPs, built on quasi-measurable structures that avoid classical measurability pathologies while preserving co-design compositionality. This enables: (i) distributional uncertainty over specifications and implementations, (ii) multi-stage adaptive design policies that condition later decisions on earlier realized outcomes, and (iii) queries that return probabilistic trade-offs and risk metrics (e.g., expected resource use, quantiles), rather than only deterministic feasible sets. This is the first framework that simultaneously handles general distributional uncertainty over compositional design models, supports multi-stage adaptivity, and remains closed under the standard co-design compositions. I-A Related work This work lies at the intersection of system-level design for CPS, compositional specification formalisms, and robust and stochastic optimization. A recurring gap in the literature is that methods with rich probabilistic and adaptive uncertainty models typically lack a modular, multiobjective, compositional language for heterogeneous subsystems, while compositional design frameworks often rely on coarse uncertainty descriptions that do not support quantitative risk or staged policies. I-A 1 System-level design in robotics and CPS System-level co-optimization of hardware and software has long been recognized as central to robotics and CPS [ 28 ] , including classical multiobjective formulations for mechanism and system design [ 19 ] and cross-layer CPS methodologies spanning sensing, computation, and control [ 23 ] . Recent work also studies explicit cost–performance trade-offs in probabilistic planning models (e.g., action-set expansion versus expected performance) [ 22 ] . These efforts motivate holistic reasoning, but typically target specific problem classes rather than providing a general compositional language that simultaneously supports heterogeneous components, multiobjective trade-offs, and distributional uncertainty. I-A 2 Contract-based and compositional design Assume–guarantee contracts enable modular reasoning by associating components with assumptions on their environment and behavioral guarantees [ 14 ] , with broad applications from circuits to autonomy [ 3 ] . They naturally support open-system composition [ 1 ] , but require substantial adaptation when subsystems interact through closed interfaces or explicit resource exchanges [ 14 , 9 ] . Moreover, contracts emphasize satisfaction/refinement of specifications, and only indirectly capture multiobjective trade-offs or quantitative risk; probabilistic extensions typically focus on satisfaction probabilities rather than compositional optimization over heterogeneous resources and functionalities. I-A 3 Robust, stochastic, and adaptive optimization Robust and stochastic optimization provide mature tools for uncertainty, including chance constraints and distributionally robust formulations [ 18 , 4 , 24 ] . Multi-stage stochastic programming models here-and-now versus wait-and-see decisions via policies over evolving uncertainty, often implemented with scenario trees [ 16 , 17 ] . Such representations can be restrictive under continuous and high-dimensional uncertainty, and many approaches focus on a single scalar objective (e.g., expected cost) rather than multiobjective trade-offs across interconnected subsystems. They also commonly abstract away how observations depend on prior design choices. Our goal is to import distributional and adaptive ideas into a compositional, multiobjective co-design setting. I-A 4 Monotone co-design and uncertainty Monotone co-design models components as monotone relations between functionalities and resources and composes them via series/parallel/feedback while preserving order structure and enabling optimization [ 28 ] . It has been instantiated across autonomous systems and control design [ 25 , 27 , 20 ] and beyond [ 26 , 21 ] . Uncertainty has primarily been treated via interval models that guarantee conservative feasibility [ 8 ] , but do not encode probabilistic risk or adaptive policies. This paper addresses these limitations by enriching monotone co-design with compositional distributional uncertainty and multi-stage adaptivity. I-B Statement of Contribution The main contributions of this paper are fourfold. First, we introduce probability measures and parametric uncertainty over DPsusing quasi-measurable spaces, yielding a distributional modeling layer that is closed under the algebra of co-design compositions. Second, we formalize multi-stage design processes in which implementations and specifications can depend on intermediate observations and earlier choices, generalizing interval uncertainty toward an adaptive stochastic-design setting while retaining compositionality. Third, we define query operators that map distributional co-design models to probabilistic performance characterizations (e.g., feasibility probabilities, distributions of minimal required resources), enabling risk-aware system design. Finally, we illustrate the framework on unmanned aerial vehicle (UAV)co-design with uncertain task profiles and component specifications, showing how distributional and adaptive modeling reveals trade-offs and robustness properties that interval methods cannot capture. I-C Structure of the paper Section II introduces notation and mathematical preliminaries. Section III reviews monotone co-design and interval uncertainty. Section IV develops the distributional and adaptive extension (distributions over DPs, implementations/specifications, and queries/observations), utilizing Quasi Universal Spaces. Section V presents the UAVcase study.

Read the complete text