Thermal Control for Swarm Robots

ISEF Category: Embedded Systems

Subcategory: Microcontrollers  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Your robot can slow down for a weird reason, heat. Push a tiny computer hard, and its speed drops to protect itself. In a swarm, that problem multiplies across every board. Your project asks if smart scheduling can keep the whole team thinking longer.

What Is It?

A thermal-management governor is a control system that watches temperature and changes how hard each computer works. Think of it like a coach rotating players so no one gets exhausted. In this project, the computers are microcontrollers, such as Pi Picos, and the workload is inference, which means making a prediction from sensor data or a model.

The tricky part is thermal coupling. That means one board warming up can affect nearby boards, especially if they sit close together or share a case. Instead of treating each board like it lives alone, you build a model that predicts how heat spreads across the swarm. Then the governor uses that model to decide which board should run which task, and when.

This topic sits at the intersection of embedded control, scheduling, and thermal modeling. You are not just measuring temperature. You are testing whether a smarter control strategy can keep total throughput higher for longer.

Why This Is a Good Topic

This is a strong science fair topic because you can change one control rule, measure a clear outcome, and compare it against a simple baseline. The real-world link is obvious, since robots, drones, and edge devices all lose performance when they get hot. You can learn control logic, data logging, model fitting, and experimental design without needing a full-size robot platform.

Research Questions

• How does thermal-aware task scheduling change sustained inference throughput compared with round-robin scheduling?
• What is the effect of board spacing on peak temperature in a Pi Pico swarm?
• Does a learned thermal-coupling model predict future board temperatures better than a single-board model?
• To what extent does workload switching reduce thermal throttling events across multiple microcontrollers?
• Which scheduling rule keeps the lowest average temperature while preserving the most completed inferences?
• How does the number of active boards affect heat buildup and throughput loss in a compact swarm?

Basic Materials

• Pi Pico boards or similar microcontrollers.
• Temperature sensors or thermistor modules.
• Breadboard and jumper wires.
• USB power supplies or powered USB hub.
• Small fans for cooling tests.
• Digital thermometer with probe.
• Laptop for code upload and data logging.
• Cardboard, foam board, or 3D-printed mounts for fixed board spacing.
• Tape measure or ruler for repeatable placement.
• Stopwatch or timestamped logging setup.

Advanced Materials

• Multiple Pi Pico boards with matching firmware.
• Infrared camera or thermal camera for surface mapping.
• Environmental chamber or controlled test enclosure.
• Current sensor modules for power draw logging.
• High-resolution temperature sensors placed at several board locations.
• Logic analyzer for timing inference cycles.
• Optional heat sinks or passive spreader materials for comparison runs.
• DAQ or microcontroller-based data acquisition system for synchronized logs.
• Mounting hardware that fixes board spacing and airflow conditions.
• Reference temperature probe for calibration checks.

Software & Tools

• Python: Cleans logs, plots temperature curves, and compares scheduling rules.
• Arduino IDE or Thonny: Uploads firmware and helps test microcontroller behavior.
• ImageJ: Measures thermal image regions if you use a thermal camera export.
• Jupyter Notebook: Organizes analysis, model fitting, and graphing in one place.
• R: Runs statistical tests and helps compare throughput across conditions.

Experiment Steps

1. Define the exact throughput metric you will optimize, such as completed inferences per unit time, and decide how you will measure temperature on each board.
2. Map the baseline behavior of your swarm under fixed workloads so you know how fast heat builds up without thermal awareness.
3. Build a simple thermal-coupling model that links one board’s workload to its own temperature and the temperatures of nearby boards.
4. Design at least one scheduling rule that uses the model to move work away from hotter boards before throttling starts.
5. Plan controls that keep hardware, spacing, airflow, and sensor placement consistent across test runs.
6. Compare the baseline and the governor with the same task mix, then analyze both sustained throughput and temperature spread.

Common Pitfalls

• Measuring only one board, which hides heat transfer between neighbors and makes the model look better than it is.
• Changing airflow between trials, which alters temperature rise and breaks fair comparisons.
• Using unmatched sensor placement on different boards, which makes some temperatures look higher just because the probe sits closer to a heat source.
• Testing with workloads that are too light, which never push the system into the thermal range where scheduling matters.
• Forgetting to log timestamps in sync across boards, which makes it hard to connect workload changes to later temperature spikes.

What Makes This Competitive

A stronger version of this project uses a clear baseline, a real predictive model, and a careful test of whether the model helps more than a simple temperature threshold. You can push it further by comparing different swarm layouts, different task mixes, or different prediction methods. Strong analysis matters here, especially if you can show when the governor helps, when it fails, and why. That kind of result feels real, not just decorative.

Project Variations

• Test the same governor on a different microcontroller family, such as ESP32 boards, to see how hardware design changes thermal behavior.
• Compare a learned thermal model with a hand-built physics-based model to see which one schedules work better.
• Use vision, sensor fusion, or tiny audio inference as the workload to test whether model complexity changes the thermal tradeoff.

Learn More

• Raspberry Pi Pico documentation: Read the official hardware and software guides, available from Raspberry Pi documentation pages.
• MIT OpenCourseWare, Feedback Control Systems: Learn control basics and stability ideas from free course notes and lectures.
• NASA NTRS: Search for thermal control papers on electronics and spacecraft systems to find modeling ideas.
• IEEE Xplore abstract pages: Search for papers on thermal-aware scheduling and edge computing, then read abstracts and accessible previews.
• PubMed: Search for review articles on thermal stress, throttling, and heat management in compact devices.