Designing a sophisticated electronic device such as a delivery drone requires selecting many options, such as choosing motors and batteries that minimize costs while maximizing the payload the drone can carry or the distance it can travel.
Solving this mystery is no straightforward task, but what happens if designers don’t know the exact specifications of each battery and motor? Moreover, the actual performance of these components will likely be affected by unpredictable factors, such as changing weather along the drone’s path.
MIT researchers have developed a fresh framework that helps engineers design sophisticated systems to explicitly account for such uncertainty. The framework allows them to model the performance trade-offs of a device with many interconnected parts, each of which may behave in unpredictable ways.
Their technique captures the probabilities of multiple outcomes and trade-offs, giving designers more information than many existing approaches, most of which typically only model best- and worst-case scenarios.
Ultimately, this framework could facilitate engineers develop sophisticated systems such as autonomous vehicles, commercial aircraft, and even regional transportation networks that are more resilient and reliable in the face of real-world unpredictability.
“In practice, the components of a device never behave exactly as you think. If someone has a sensor whose operation is uncertain, the algorithm is uncertain, and the design of the robot is also uncertain, now they have a way to combine all of these uncertainties to be able to come up with a better design,” says Gioele Zardini, Rudge and Nancy Allen Postdoctoral Fellow in the Department of Civil and Environmental Engineering at MIT, and principal investigator in the Systems Laboratory Information and Decision Sciences (LIDS), in a department affiliated with the Institute for Data, Systems and Society (IDSS), and senior author of the book paper on this framework.
In the article, Zardini is joined by lead author Yujun Huang, an MIT graduate student; and Marius Furter, a graduate of the University of Zurich. The results of the study will be presented at the IEEE Conference on Decision and Control.
Given the uncertainty
The Zardini group researches co-design, a method of designing systems composed of many interconnected components, from robots to regional transport networks.
In previous work, researchers modeled each component of a collaborative design without taking uncertainty into account. For example, the performance of every sensor designers could choose for a drone has been improved.
However, engineers often do not know the exact performance specifications of each sensor, and even if they do, it is unlikely that the sensor will match the specifications perfectly. At the same time, they don’t know how each sensor will behave when integrated into a sophisticated device, or how performance will be affected by unpredictable factors such as weather.
“With our method, even if you’re not sure what the specifications of your sensor will be, you can still design the robot to maximize the results you want,” Furter says.
To achieve this, researchers incorporated the concept of uncertainty into existing category theory frameworks.
Using some mathematical tricks, they simplified the problem into a more general structure. This allows them to employ the tools of category theory to solve co-design problems in a way that takes into account a range of uncertain outcomes.
By reframing the problem, researchers can capture how multiple design choices influence each other, even when their individual performance is uncertain.
This approach is also simpler than many existing tools, which typically require extensive domain expertise. Thanks to the plug-and-play system, you can change the arrangement of system components without violating any mathematical constraints.
And because no specialized knowledge in a given field is required, the platform can be used by a multidisciplinary team, each member of which designs one component of a larger system.
“Designing an entire UAV is not feasible for just one person, but designing a component of it is doable. By providing a framework for these components to work together in an uncertainty-aware way, we have made it easier for people to evaluate the performance of the entire UAV system,” Huang says.
More detailed information
Scientists used this fresh approach to select perception systems and batteries for the drone that would maximize its payload while minimizing cost and weight over its lifetime.
While each perception system may offer different detection accuracy in different weather conditions, the designer doesn’t know exactly how its performance will vary. This fresh system allows the designer to account for these uncertainties when considering the overall performance of the drone.
Unlike other approaches, their framework reveals clear advantages of each battery technology.
Their results show, for example, that at lower loads, nickel-metal hydride batteries provide the lowest expected operating cost. It’s impossible to fully capture this knowledge without taking uncertainty into account, Zardini says.
While the other method can only show best- and worst-case scenarios for lithium-polymer battery performance, its framework provides the user with more detailed information.
For example, it shows that if the payload of the drone is 1750 grams, there is a 12.8% chance that the battery design will be unfeasible.
“Our system provides trade-offs and then the user can justify the design,” he adds.
In the future, researchers want to improve the computational efficiency of their problem-solving algorithms. They also want to extend this approach to situations where the system is designed by multiple collaborating and competing parties, such as a transport network where railway companies operate using the same infrastructure.
“As systems increase in complexity and incorporate increasingly disparate components, we need a formal framework for designing these systems. This paper shows how to compose large systems from modular components, understand design trade-offs, and, importantly, do so with the concept of uncertainty. This provides an opportunity to formalize the design of large-scale systems with components that enable learning,” he says Aaron Ames, Bren professor of mechanical and civil engineering, Control and Dynamical Systems and Aerospace at Caltech, who was not involved in this research.