Automated Insulin Delivery (AID) Algorithms

Introduction to Automated Insulin Delivery (AID) Algorithms

An automated insulin delivery (AID) system connects a continuous glucose monitor (CGM) to an insulin pump through software that decides how much insulin the pump should deliver. That software is the algorithm. The FDA calls these devices artificial pancreas device systems and notes that they are also known as closed-loop or automated insulin delivery systems [1]. The "artificial pancreas" name oversells them a little. The systems we read about are hybrids: the algorithm runs background insulin on its own, but the user still has to tell it about meals [2][3][4][5].

How AID Algorithms Work

Every few minutes, the algorithm takes the latest CGM reading, estimates where glucose is heading, and changes insulin delivery to match [3][4]. Tandem's Control-IQ, for example, predicts glucose 30 minutes ahead and can adjust or suspend basal (background) insulin every 5 minutes [3]. A 2021 review in Diabetologia groups the control methods into three families [2]:

• Model predictive control (MPC) uses a mathematical model to predict glucose over a set time window, then picks the insulin delivery that keeps the prediction closest to target [2].
• Proportional-integral-derivative (PID) control looks at how far glucose is from target, how long it has been off target, and how fast it is changing [2].
• Fuzzy logic adjusts insulin with approximate rules that encode what experienced diabetes clinicians would do [2].

The same review lists the MiniMed 670G and 780G as PID systems with insulin feedback and Control-IQ as a "treat to range" predictive algorithm [2]. Insulet's algorithm is an MPC design. It began as a personalized MPC algorithm developed at the University of California, Santa Barbara, and in a 2018 feasibility study it chose a dose every 5 minutes to keep predicted glucose over the next 60 minutes close to target [22]. Some designs go further in adapting to the person. Beta Bionics' iLet software, for instance, starts from body weight alone and keeps adapting to the user's insulin needs over time [5]. Researchers are also testing machine learning: one 2022 study built a deep-learning framework to detect and estimate meals in artificial pancreas systems [6]. These design choices show up in daily use. Control-IQ's glucose targets can't be customized and change only when the user turns on sleep or exercise mode [3]. Control-IQ+ keeps that approach, with separate Normal, Sleep and Exercise modes that each have their own target range [23]. The 780G, by contrast, added adjustable targets and automatic correction boluses over the 670G [2]. For a closer look at the two main approaches, see our article on PID vs. MPC algorithms.

Where the Algorithms Came From

The clearest paper trail we found runs from a university lab to a commercial pump. According to the Control-IQ pivotal trial report, the system is a third-generation descendant of DiAs, a mobile closed-loop system developed at the University of Virginia and later implemented as inControl by TypeZero Technologies [7]. Tandem signed a license agreement with TypeZero in July 2016 [8], and Dexcom acquired TypeZero in August 2018 [20]. The other pump makers run their own algorithms: MiniMed's SmartGuard [9][11], Insulet's SmartAdjust [4] and Beta Bionics' iLet dosing software [5].

Manufacturers and Product Lines

Key manufacturers in the US AID market, as of October 2026, include:

• MiniMed (Medtronic's diabetes business, now listed on Nasdaq as MMED): Its MiniMed 780G, approved by the FDA in April 2023, and its MiniMed Flex, cleared in March 2026 and launched in the US on June 17, 2026, both run the SmartGuard algorithm [9][10][11]. Flex is cleared for type 1 diabetes from age 7 and for adults with insulin-requiring type 2 diabetes [9]. The MiniMed 670G, the first hybrid closed-loop system the FDA approved (September 28, 2016), is the earlier generation; the 780G was developed as its second-generation successor [1][2].
• Tandem Diabetes Care: The t:slim X2 and Tandem Mobi pumps run Control-IQ+, which Tandem launched in the US in March 2025 and calls the latest generation of its advanced hybrid closed-loop algorithm [21][12][13]. Control-IQ+ is indicated for type 1 diabetes from age 2 and type 2 diabetes from age 18 [13]. Control-IQ, which the FDA authorized in December 2019, is the earlier version [3][21]. Tandem's Basal-IQ technology doesn't appear on its current t:slim X2 or Mobi pages, and we couldn't confirm its current status [12][13].
• Insulet: The Omnipod 5 runs SmartAdjust technology on the Pod itself, adjusting insulin every 5 minutes from sensor readings [4][14]. The FDA cleared SmartAdjust in August 2024 for type 1 diabetes from age 2 and type 2 diabetes in adults [4]. Omnipod 5 works with the Dexcom G7 and G7 15 Day sensors and with the FreeStyle Libre 2 Plus and Libre 3 Plus sensors [14].
• Beta Bionics: The iLet Bionic Pancreas was cleared in May 2023 for type 1 diabetes from age 6 [5]. It doesn't use carb counts. Users announce a meal and describe it as "Usual," "More" or "Less" than usual for that meal [5]. The iLet works with Dexcom CGMs and the FreeStyle Libre 3 Plus sensor [15].
• Sequel Med Tech: The twiist AID System runs the twiist Loop algorithm, intended for type 1 diabetes from age 6. It adjusts or suspends basal insulin automatically and delivers correction boluses after the user confirms them [16]. twiist works with the FreeStyle Libre 3 Plus sensor and the implantable Eversense 365 [16].
• Bigfoot Biomedical: Abbott completed its acquisition of Bigfoot on September 22, 2023 [17]. Bigfoot's FDA-cleared product, Bigfoot Unity (May 2021), pairs an Abbott CGM sensor with caps for insulin pens and is classified as not for use with AID systems [18]. We found no FDA record of a Bigfoot pump-based AID system, so the status of that work is unclear.

Product Comparison

The trials we read measure success mainly by time in range (TIR), the share of the day that glucose stays inside a target band, along with time spent low [2][7]. In Control-IQ's six-month randomized trial of 168 people aged 14 to 71, time in range rose from 61% to 71% with the closed-loop system and stayed at 59% with a standard sensor-augmented pump [7]. In the FLAIR crossover trial in adolescents and young adults, MiniMed's advanced hybrid closed-loop system (the 780G algorithm) gave more time in range than the 670G with similar time in hypoglycemia, and users stayed in automatic mode longer (86% vs. 75% of the time) [2]. Those trials used different designs and comparison groups, so their numbers shouldn't be compared directly. The American Diabetes Association's 2026 Standards of Care call AID the preferred insulin delivery method in type 1 diabetes and in type 2 diabetes treated with multiple daily injections. They also cite psychosocial benefits and a review that found AID systems cost-effective [19].

Pitfalls, Warnings, and Issues

The FDA's review of Control-IQ is a useful map of what can go wrong [3]:

• Technical Issues: The FDA lists risks from latent software flaws and from lost communication with connected devices. Insecure data transmission and wireless interference are on the same list [3]. In the Control-IQ pivotal study, one case of diabetic ketoacidosis (DKA) followed an infusion set failure. Two software problems temporarily suspended the study until patches fixed them [3].
• Sensor Dropouts: The algorithm needs a steady stream of CGM readings. When sensor data are missing or unusable, for example after a lost wireless connection, the 780G delivers a preset "safe basal" rate for up to 90 minutes and then switches to Manual Mode if the problem doesn't clear [11]. The iLet lets the user enter fingerstick readings to keep dosing, but only for a limited time [5].
• Meal Spikes: Insulin given under the skin is absorbed more slowly than food is digested, so the glucose rise after a meal is still hard for these systems to control [2].
• User Burden: Calibration is less of a chore than it used to be. The 670G's Guardian Sensor 3 needed about four to six fingersticks a day [2]. Even on the 780G, that older sensor needed a calibration at least every 12 hours, while the newer Guardian 4 sensor needs none [11]. The routine work now is replacing sensors (the Guardian 4 lasts up to seven days) and telling the system about meals [3][5][11].
• Cost and Accessibility: The ADA notes that US pump therapy is more common among people with higher socioeconomic status, as reflected by private health insurance, income and education [19]. Problems with cost are among the reasons people stop pump therapy, and the ADA says closing these gaps in access may lower health disparities [19].

Conclusion

As of October 2026, US AID algorithms come from at least five companies: MiniMed's SmartGuard, Tandem's Control-IQ+, Insulet's SmartAdjust, Beta Bionics' iLet software and the twiist Loop algorithm [4][5][9][12][16]. The ADA calls AID the standard of care for people with type 1 diabetes who can use it safely [19]. Which system fits a given person, and how it is set up, is a decision to make with your care team.