Get started

Case studies

Five deployments that put SENTIL on a real workload, from Autonomous Vehicle runtime monitoring on CARLA to a real drive on a Raspberry Pi to glucose monitoring.

Each of the five studies below takes a real-world system, writes its safety property, runs the monitor, and reports the robustness.

The case studies are in two types. Three run offline on a recorded or simulated trace and reproduce on plain CPU with no special hardware: the CARLA drive, the insulin controller, and the circadian network. Two run the monitor live inside a target platform, watching signals as they arrive: the Raspberry Pi supervisor and the Apollo Cyber RT component. AUTOSAR Adaptive monitoring runs the same kind of live study over SOME/IP; that one is written up on the AUTOSAR language guide.

The studies

Reproducing them

Every case study has a directory under experiments/ with the code to rerun the study and produce the numbers.

pip install sentil numpy matplotlib
python experiments/glucose_control/glucose_control.py
python experiments/circadian_gene_network/circadian_gene_network.py
python experiments/carla_driving/monitor_drive.py --trace experiments/carla_driving/results/drive.json

The claims document at docs/CLAIMS.md records the expected value and tolerance for each artifact, and python scripts/check_claims.py fails the build if any reproduced number drifts outside its bound. The Raspberry Pi experiment needs a board, and the Apollo study needs its Cyber RT runtime; each page states what it needs and what to expect where the hardware is absent.

Edit this page on GitHub