Choosing the right control algorithm: PID, MPC, and hybrid approaches

The control algorithm is the defining engineering decision in any closed-loop insulin delivery system, and the field has converged on three dominant paradigms: PID-based control, Model Predictive Control (MPC), and hybrid compound architectures that combine both. Each offers a different balance of computational simplicity, predictive power, and patient-adaptability — and the patent record from 2006 to 2026 makes clear that no single approach dominates across all clinical scenarios.

PID and risk-space PID (rPID)

Traditional PID controllers are reactive: they compute insulin delivery from the proportional, integral, and derivative terms calculated from the error between measured glucose and a target setpoint. Medtrum Technologies has extensively patented a risk-transformed variant called rPID, in which asymmetric blood glucose values in the raw physical space are mapped into an approximately symmetric risk space before control computation. This rPID approach retains the simplicity and robustness of classical PID while adding the precision advantages of risk-space transformation. The same risk-space concept is extended to MPC in corresponding rMPC filings.

Model Predictive Control (MPC) and zone MPC

MPC is the more computationally sophisticated approach. It explicitly uses a forward-looking physiological model to predict future glucose trajectories and optimise the insulin infusion rate over a prediction horizon, subject to safety constraints. Diabeloop’s MPC-based controller determines a maximum allowable insulin injection amount and generates a delivery control signal based on this limit and the current required quantity. The University of California Regents implemented zone MPC with daily periodic target-zone modulation, striving to maintain an 80–140 mg/dL glucose zone during the day and a 110–220 mg/dL zone at night, with a 2-hour smooth transition between them.

Hybrid compound algorithms: the current frontier

Hybrid compound control represents the most advanced algorithmic approach in the current patent record. Medtrum Technologies has patented a family of hybrid artificial pancreas algorithms that deeply combine PID and MPC: in the cPID variant, the input to the PID stage is the intermediate output of MPC; in the cMPC variant, the input to MPC is the output of the PID controller. A further compound artificial pancreas algorithm calculates two independent insulin infusion amounts (I1 and I2) from independent algorithm stages and then optimises them to produce a final infusion amount (I3).

“Hybrid PID-MPC compound algorithms that cross-couple the intermediate output of each algorithm as the input to the other outperform either algorithm independently across diverse patient scenarios.”

Additional algorithmic approaches in the patent record include autoregressive models with explicit insulin absorption lag factors (Medtrum Technologies, 2018), Iterative Learning Control (ILC) using a run-to-run policy to progressively update long-acting insulin dosing plans against sparse glucose measurement data (Harvard University, 2021), and closed-loop adaptive control by glycaemia model inversion (Universidad Politecnica de Madrid, 2010). The University of California also contributed individualized controller schemes combining IMC-PID and MPC with personalized insulin feedback (IFB), calibrated using a priori patient basal insulin data.

Insulin-on-Board management and pharmacokinetic modelling

Accurate Insulin-on-Board (IOB) tracking is safety-critical in any closed-loop insulin delivery system: over-delivery relative to actual IOB is the primary mechanism of algorithmic hypoglycaemia. The technical challenge is compounded by the physical limitation that subcutaneous insulin absorption cannot exceed a maximum rate — meaning naive IOB calculations that ignore depot saturation will systematically underestimate active insulin after large bolus deliveries.

Subcutaneous absorption limits and depot saturation

Insulet Corporation addresses the subcutaneous absorption saturation problem directly: the system calculates an aggregate IOB amount over a first time period, determines whether this exceeds a maximum depot formation threshold, and carries a residual forward to increase the subsequent individual IOB estimate. This approach prevents under-counting of effective IOB when large depot amounts are present. Bigfoot Biomedical takes a complementary relative IOB approach: a processor computes a reference IOB value representing what a standard open-loop regimen would have delivered, and an automated IOB value representing what the closed-loop system has automatically dispensed, with the difference (relative IOB) used to guide subsequent dosing decisions.

Insulet Corporation has also developed dynamic on-board insulin profile customisation, adapting IOB decay curves based on individual recent glucose history rather than population-average static profiles. This moves the field away from one-size-fits-all pharmacokinetic assumptions toward individually calibrated absorption modelling.

Patient digital twins and adaptive PK/PD modelling

Pharmacokinetic/pharmacodynamic (PK/PD) modelling is embedded directly into closed-loop control in the patient digital twin approach by Medtronic MiniMed. Sensor glucose and meal data from recent days are used to fit a user-specific PK/PD model, which then computes updated controller parameters — enabling the controller to adapt to individual patient physiology over time. The Trustees of Boston University implemented an adaptive MPC in which a linear empirical input-output patient model is recursively updated online based on glucose measurements, minimising both subcutaneous insulin accumulation and control aggressiveness in an augmented objective function.

Total Daily Dose tracking as an insulin sensitivity proxy

Total Daily Dose (TDD) tracking is used as a proxy for insulin sensitivity in multiple systems. Eli Lilly has patented a two-phase filtered TDD tracking method, in which an initial tracking phase applies one set of rate limits and a steady-state phase applies a different set, allowing the system to converge on a reliable TDD estimate and use it to set system gains. Medtrum Technologies integrates motion sensing directly into TDD and current infusion algorithms, using an onboard motion sensor to detect physical activity status as a variable factor that automatically accounts for exercise-induced changes in insulin sensitivity.

Hardware architecture: from multi-device to single-insertion platforms

The physical architecture of a closed-loop drug delivery system is as consequential as its algorithm. Systems range from loosely coupled multi-device configurations — where a separate CGM, controller, and pump communicate wirelessly — to fully integrated wearable platforms that co-locate sensing and infusion in a single subcutaneous insertion point. The patent record shows a clear directional trend toward integration.

Single-insertion integrated sensing and infusion

Medtrum Technologies has developed intelligently controlled miniature fully closed-loop artificial pancreas systems in which the infusion cannula itself serves dual roles: as the drug delivery channel and as an electrode for blood glucose sensing. The cannula wall electrodes and conductive-area electrodes allow analyte detection and insulin infusion from a single subcutaneous insertion point. The corresponding EP-jurisdiction grant confirms active status for this integrated design as of 2026. This single-insertion approach reduces infection risk and patient burden compared with configurations requiring separate CGM and infusion sites.

Mechanical infusion drive mechanisms

Medtrum Technologies has patented two distinct mechanical drive approaches. The unilaterally driven closed-loop artificial pancreas uses a single-direction linear actuator and reset unit to precisely advance a piston-screw mechanism. The bilaterally driven variant applies forces in two directions, enabling finer-resolution delivery steps and higher dosing efficiency. The choice between these architectures involves trade-offs between mechanical simplicity and delivery resolution that will be familiar to any medical device engineer working on miniaturised infusion systems.

Embedded firmware optimisation for resource-constrained processors

Insulet Corporation addresses the challenge of running MPC on wearable processors in firmware. The method uses a stepwise coarse-to-fine search across a discrete drug delivery amount space: a coarse pass narrows the solution range, and a refined pass identifies the cost-minimising delivery quantity, where the cost function evaluates predicted glucose and insulin trajectories using a recursive glucose model. This embedded optimisation approach avoids the need for continuous real-time convex optimisation solvers — a critical constraint for battery-powered wearable devices. According to IEEE standards for embedded medical device software, resource-constrained optimisation is a recognised design challenge in Class III implantable and wearable systems.

Dual-drug systems and multi-module redundancy

Fully closed-loop dual-drug systems capable of delivering both insulin (hypoglycaemic) and glucagon (anti-hypoglycaemic) have been patented by Medtrum Technologies. This architecture uses independent infusion modules for each drug class, both governed by a shared algorithm-driven program module that determines which drug to administer based on the current blood glucose trajectory. Separately, Medtrum Technologies has also patented a distributed control architecture in which detection, infusion, and electronic modules each contain their own embedded control units, with different modules assuming control priority based on operational conditions — preventing single-point failure from compromising insulin delivery.

Dexcom has addressed the integration of a continuous glucose sensor with an insulin delivery device and controller module, supporting manual, semi-automated, and fully automated operation. Becton Dickinson described an earlier foundational architecture in which a sensor system, feedback/model-based controller, and intradermal insulin infusion device constitute the system, with time-delay reduction as a key design objective — particularly relevant because intradermal delivery reduces pharmacokinetic lag relative to subcutaneous delivery, as noted by FDA guidance on closed-loop insulin delivery device classification.

Safety constraints, dosing limits, and fault management

Safety enforcement is a non-negotiable engineering requirement in closed-loop insulin delivery: over-delivery can cause life-threatening hypoglycaemia within minutes. The patent record reveals a layered safety architecture in which multiple independent mechanisms each provide a backstop against dangerous over-infusion, and the failure of any single mechanism does not compromise the patient.

Maximum infusion rate enforcement

Medtronic MiniMed has extensively patented methods for calculating user-specific insulin infusion upper limits based on fasting blood glucose values, total daily insulin, and insulin delivered during fasting periods. When the controller-computed infusion rate exceeds the patient-specific maximum, the system substitutes a rate capped at that limit. LifeScan IP Holdings similarly addresses maximum dosing from a bolus-rate computation standpoint, computing a maximum delivery rate from default basal, temporary basal, and extended bolus rate data. Both approaches reflect the principle — also codified in ISO 11608 standards for needle-based injection systems — that software-enforced upper limits must be independent of the primary control algorithm.

Closed-loop entry safeguards and dynamic startup setpoints

Closed-loop entry safeguards govern when a system transitions from open-loop to closed-loop operation. Medtronic MiniMed’s safeguarding techniques require analysis of CGM calibration factors and corresponding timestamp data before the system is permitted to enter closed-loop mode. The same patent family details a PID-IFB (Insulin Feedback) closed-loop algorithm with IOB compensation running as an additional safety overlay. Target glucose setpoint management during closed-loop startup is critical because a large gap between the current glucose level and the target setpoint can cause aggressive initial dosing. Medtronic MiniMed addresses this with a dynamic final target glucose value that decreases gradually toward the setpoint during startup.

Risk-based delivery transformation

DexCom’s approach moves safety enforcement into the algorithm’s objective function itself. A model-match evaluator quantifies the degree to which recent glucose measurements are inconsistent with recent insulin — a discrepancy indicator for potential sensor error or patient metabolic disturbance — and then adjusts insulin delivery rate based on quantified hyper- or hypoglycaemic risk. This approach, disclosed in a 2026 filing, represents the most recent evolution of risk-aware closed-loop control in the dataset.

Patent landscape: key players and the six innovation trends shaping the field

The closed-loop insulin delivery patent landscape spans more than 50 filings across WO, EP, US, CN, JP, KR, AU, RU, ES, TW, and CA jurisdictions, with filings from 2006 to 2026 reflecting sustained global commercial interest. Understanding who is filing what — and in which direction the field is moving — is essential for R&D engineers and IP professionals making technology investment decisions.

Dominant assignees

• Medtrum Technologies Inc. is by far the most prolific in this dataset, with over 15 distinct filings covering rPID/rMPC algorithms, hybrid compound algorithms, integrated single-insertion hardware, bilateral/unilateral drive mechanisms, multi-drug dual infusion, motion-sensor integration, and multi-module redundant architectures. Filings span 2018 to 2026, with active status in EP and WO jurisdictions.
• Medtronic MiniMed contributes major innovations in closed-loop safeguarding techniques, maximum infusion rate enforcement, PID-IFB control algorithms, dynamic target setpoint management, and patient digital twin-based PK/PD parameter adaptation.
• Insulet Corporation focuses on IOB management (specifically subcutaneous absorption rate limits), embedded firmware dose optimisation using coarse-to-fine search, and dynamic personalised IOB profile customisation.
• The Regents of the University of California pioneered zone MPC and time-varying daily target range modulation, establishing the mathematical framework used by several commercial implementers.
• Harvard University and Boston University represent the academic frontier, with iterative learning control using sparse measurements and adaptive recursive MPC with augmented objective functions respectively. Academic contributions to this field are tracked by WIPO as part of its global health technology patent monitoring programme.

Six innovation trends visible in the patent record

1. Increasing integration of sensing and infusion hardware to a single-insertion platform
2. Shift from static population-average pharmacokinetic parameters toward individually adaptive patient models (digital twins)
3. Risk-space transformation of glucose variables to improve asymmetric control
4. Dual-drug (insulin + glucagon) fully closed-loop systems
5. Embedded optimisation (coarse-to-fine search) enabling MPC on wearable processors
6. Digital twin personalisation of controller parameters using recent sensor and meal data