Safe Reinforcement Learning for climate control
DQN-based supervisory control and physical deployment.
Related research & publications →Toba, Indonesia · Indoor farming research
AI-Driven Indoor Farming: From Sensor Data to Intelligent Control
Exploring safe reinforcement learning, 3D microclimate mapping, VPD regulation, and energy-aware control in an instrumented vertical farm.
3D spatial microclimate mapping
Sparse sensors, Kriging, and residual correction.
Related research & publications →Data-driven and energy-aware indoor farming
IoT, edge control, VPD, and operational evidence.
Related research & publications →
- 10 ft
- Instrumented research container
- 27 + 1
- Indoor and outdoor sensor nodes
- Toba
- North Sumatra, Indonesia
Weather & microclimate
Outside in Toba. Inside the container.
BMKG forecasts for Balige I, Toba Regency, alongside a demonstration of container microclimate. Two distinct sources, with explicit data provenance.
BMKG · Official forecast
Balige I · Toba
Three-hour weather forecasts; not live measurements or long-term climate statistics. Balige I does not represent all of Toba.
Loading BMKG forecast…
SYNTHETIC DATA · DEMONSTRATION
10-ft container microclimate
All values and curves below are mathematically generated for illustration. They are not sensor readings, experimental results, or BMKG data. Simulated lighting: 06:00–22:00 WIB. Light is expressed in lux, not PPFD.
Air temperature
— °C
Simulated
24 h range: —
Relative humidity
— % RH
Simulated
24 h range: —
Light intensity
— lux
Simulated
24 h range: —
CO₂ concentration
— ppm
Simulated
24 h range: —
Curves: past 24 hours, left → right; each metric uses its own scale. Simulation updates every minute. Simulation time: —
Research stories
What can one container teach us?
Spatial microclimate
The AC cooled the farm—but not uniformly
Across 544 synchronized observations, average temperature fell while spatial spread increased. Explore what this event reveals about evaluating cooling with multiple sensors.
Read the story →
Climate control
Good temperature is not enough
A 48.1-hour observation revealed nighttime humidity limitations. Read the VPD findings and their limits.
Read the story →
IoT · Edge · Cloud
From sensor readings to local decisions
How 27 sensor nodes, a Jetson Nano, and a local dashboard connect sensing, control, and remote monitoring.
Read the story →About
Applied research in one measurable facility
The platform combines hydroponic cultivation, artificial lighting, distributed temperature–humidity sensing, CO₂ monitoring, actuator telemetry, and energy measurement.
Our work focuses on evidence: where conditions differ, how quickly the system recovers, which nodes remain outside target, and whether control improvements justify their energy cost.
27 + 1
indoor and outdoor sensor nodes
10 ft
instrumented container
20–24°C
thermal evaluation band
Facility
From sensing to actuation
Internal dimensions are approximately 2.34 × 2.82 × 2.39 m. Hydroponic racks line both sides of the aisle, with artificial lighting, a 1-PK split AC, humidification, nutrient management, and CO₂ enrichment.
Sensing
Distributed temperature, humidity, light, CO₂, weather, and energy measurements.
Processing
Edge and cloud processing for synchronization, diagnostics, and control.
Storage
Time-series records and dashboards for traceable operational analysis.
Actuation
Cooling, humidification, lighting, nutrients, and CO₂ under supervised control.
Documentation
Facility, crops, and real monitoring data
Instrumented facility interior
Hydroponic racks, distributed nodes, and nutrient and actuation subsystems.
Pakchoi on hydroponic racks
Pakchoi cultivated on multi-level gullies under artificial lighting.
Monitoring dashboard, July 2026
Temperature, humidity, CO₂, lighting, actuator state, current, and power recorded during operation.
Articles & findings
Latest insights
28 Oktober 2026
Bangkit Penelitian Indonesia: Semangat Sumpah Pemuda
Refleksi Kangabet tentang pemuda, integritas akademik, kemandirian ilmu pengetahuan, dan inovasi yang berdampak.
Read articleEquipment
IoT & Smart Farming Equipment: 7 practical tutorials
ESP32, MQTT, Grafana, Jetson Nano, Cloudflare, microclimate sensors and hydroponic monitoring with honest equipment guidance.
Read articleServices & Training
IoT system reviews and practical workshops
Defined deliverables, pilot pricing, and documented speaking and training experience.
Read articleSpeaking · 2024
Cybersecurity lessons from the BSSN National Cyber Exercise #9
A retrospective on malware-stealer prevention and digital security awareness, with the original IT Del report.
Read articleGovernance · 2024
Information security governance in North Sumatra
A retrospective on the November 2024 FGD covering governance, SIEM, and the NIST Cybersecurity Framework.
Read articleSpace & time
Mapping Space, Forecasting Time: Kriging and SARIMA for Indoor Farming
Kriging, IDW, RBF, ARIMA, and SARIMA: their roles, seasonal periods, and validation without data leakage.
Read articleAI & farming
Nine AI Trends Shaping Indoor Vertical Farming in 2026
A source-based overview of safe RL, digital twins, edge AI, and emerging research opportunities.
Read articleMonitoring
Grafana for Indoor Farming Monitoring
Connect sensor readings, actuator states, and operational diagnosis in one dashboard.
Read articleHaze & crops
When Smoke Hides the Sun: What Toba’s Haze Means for Farmers
Which crop stages need attention, what smoke changes, and how farmers can document risks without guessing harvest losses.
Read article27 nodes
From IoT to Edge-to-Cloud: A Resilient Architecture for Container Farming
Sensor data reaches a Jetson Nano for local decisions, while a local server and Cloudflare Tunnel make monitoring available online.
Read article4.6×
Why Do Horizontal Temperature Differences Matter More Than Rack Height?
During one AC recovery event, the horizontal temperature gap reached 3.33°C, compared with 0.72°C vertically.
Read article5.03°C
The AC Cooled the Container Farm—but Not Uniformly
Across 544 synchronized observations, average temperature fell while spatial temperature spread increased.
Read article48.16%
VPD in a Container Farm: Good Temperature Is Not Enough
A 48.1-hour observation exposed nighttime humidity limitations that average temperature alone did not reveal.
Read articleResearch
Research directions
Spatial microclimate mapping
Sparse-sensor interpolation, variogram sensitivity, residual correction, and spatial compliance.
Safe climate control
Rule-based, PID, DQN, and safety-aware supervisory control for temperature and humidity.
Thermal resilience
Fault detection, controller recovery, hotspot persistence, and post-shutdown rebound.
Energy-aware operation
Joint assessment of climate compliance, actuator power, switching, and recovery time.
Publications & community engagement
From research to community application
Explore publications in precision agriculture, sensor networks, automation, and system security, alongside community programmes with farmers and schools.
9
International journals ↗
23
International conference proceedings ↗
9
National journals ↗
6
Community programmes ↗
Resources & academic work
Collaborate with Tobafarm
Bring a question. Build a testable plan.
Choose the path closest to your needs. Scope, facilities, timelines, and deliverables are discussed before work begins.
01 · For researchers
Develop a joint study
Discuss experimental questions, sensor data, microclimate methods, and control evaluation.
Discuss joint research →02 · For industry & government
Scope a practical pilot
Explore sensing, monitoring, microclimate audits, and control prototypes with measurable outcomes.
Explore pilot support →03 · For students & educators
Learn through real systems
Find teaching materials, explore supervision, or discuss focused technical training.
Explore learning resources →