glucosim

Gymnasium Environments for Reinforcement Learning in Glucose Management

Three Gymnasium-compatible RL environments for Type 1 diabetes glucose management: basal rate optimization, meal bolus dosing, and full closed-loop insulin delivery control. Includes the Bergman minimal glucose-insulin model, Dalla Man gut absorption dynamics, a CGM sensor noise model, 30 virtual patients across three age groups (child, adolescent, adult), configurable difficulty tiers, heuristic clinical baselines, PPO RL agents, and a five-tier benchmark suite. 120 tests, MIT licensed.

Installation

pip install glucosim              # Core (numpy, scipy, gymnasium)
pip install glucosim[train]       # + SB3, PyTorch for RL training
pip install glucosim[all]         # Everything

Development install:

git clone https://github.com/HassDhia/glucosim.git
cd glucosim
pip install -e ".[all]"

Quick Start

import gymnasium as gym
import glucosim

env = gym.make("glucosim/BasalControl-v0")
obs, info = env.reset(seed=42)
for _ in range(100):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()
env.close()

Environments

Architecture

GlucoSim implements a modular simulation stack:

• Bergman Minimal Model - Three-equation glucose-insulin dynamics with RK4 integration
• Dalla Man Gut Model - Two-compartment carbohydrate absorption producing glucose appearance rates
• CGM Sensor Model - First-order lag filter with configurable Gaussian noise
• Virtual Patients - 30 patients across child/adolescent/adult groups with +/-20% parameter variability
• Baseline Agents - Random, proportional basal controller, and ICR bolus calculator
• PPO Agent - Stable Baselines3 PPO with tuned hyperparameters
• Benchmark Suite - Five difficulty tiers per environment (easy-adult to hard-child)

Paper

The accompanying paper is available at:

• PDF (GitHub)

Citation

If you use glucosim in your research, please cite:

@software{dhia2026glucosim,
  author = {Dhia, Hass},
  title = {GlucoSim: Gymnasium Environments for Reinforcement Learning in Glucose Management},
  year = {2026},
  publisher = {Smart Technology Investments Research Institute},
  url = {https://github.com/HassDhia/glucosim}
}

License

MIT License. See LICENSE for details.

Contact

Hass Dhia - Smart Technology Investments Research Institute

• Email: partners@smarttechinvest.com
• Web: smarttechinvest.com/research