Engineers rarely build under perfect information. Components don’t always perform exactly as specs promise. Environmental conditions, interactions, and unpredictable fluctuations further cloud reality. A new framework developed at MIT tackles this head-on, letting designers explicitly weave uncertainty into models of complex systems, tells MIT News.
The key context is co-design: when many subsystems must be chosen and tuned together. Think drones, autonomous vehicles, or regional transportation networks. Each part, i.e., sensors, motors, and control logic, carries uncertainty. The new method allows these uncertainties to interact through the design, rather than forcing crude worst-case versus best-case assumptions.
Behind the framework is category theory. The authors reframe design tradeoffs in a modular, composable format so that uncertainty is built into the mathematics, not tacked on later. The result: a plug-and-play approach where subsystem models can be combined without violating consistency constraints.
In practice, the team tested it by designing a drone. They considered different sensors, battery types, and payload configurations, all with unpredictable performance. The framework enabled them to optimize for expected payload, cost, or weight, while quantifying the risk of failure or infeasibility under uncertainty. For example, for a 1,750-gram payload mission, the model computed that there is a 12.8% chance a chosen battery design would fail—insight impossible with rigid deterministic models.
This method simplifies design for interdisciplinary teams. You don’t need deep domain expertise in every component; you need a coherent system that composes them under uncertainty.
The future challenges include scaling the approach, improving algorithmic efficiency, and applying it to large systems designed by multiple stakeholders, such as transport networks where different operators share infrastructure.
By embedding uncertainty into system design rather than treating it as an afterthought, this new framework could lead to more resilient, reliable, and realistic engineered systems.