Skip to content

Swarm Robotics Cluster — Design Report

Platform: Raspberry Pi Pico W (WiFi + Bluetooth) Concept: One master robot broadcasts its heading; follower robots independently steer to match it.

1. Core Concept

Rather than tracking the master's position (which requires localization and is hard to get right), each follower matches the master's heading. The master broadcasts its current heading over the network at a fixed rate. Each follower reads its own gyroscope + magnetometer, computes its own heading, and runs a closed-loop controller to steer toward the broadcast target. This makes "follow the leader's turns" a heading-synchronization problem rather than a full swarm-positioning problem — much easier to build and debug in a classroom setting.

2. Hardware Per Robot

  • Raspberry Pi Pico W (RP2040 + CYW43439 radio)
  • 9-DOF IMU (see comparison below)
  • Motor driver: TB6612FNG (preferred over L298N — more efficient, cleaner PWM)
  • Two DC gear motors + wheels + caster, simple chassis
  • LiPo battery + buck/boost regulator for the 5V logic rail

Note: the Pico W's WiFi and Bluetooth both run on the CYW43439 chip, but Bluetooth (BLE) support depends on the SDK/firmware build in use — confirm the .uf2 build supports BLE if the class wants to experiment with it later.

3. Communication: WiFi UDP Broadcast

Simplest approach for a first swarm project:

  • Master Pico W runs in Access Point mode, creating its own local WiFi network.
  • Master broadcasts small UDP packets (e.g. {"heading": 137.4, "speed": 0.6}) at a fixed rate — 20–50 Hz is plenty.
  • Followers join the AP, open a UDP socket, and just listen — no handshaking or reconnect logic needed.
  • Dropped packets don't break anything — the next one arrives 20–50 ms later.

BLE advertising (broadcast-only, no pairing) is a reasonable alternative to compare later, but UDP is easier to get working first.

4. Control Loop (same code on every robot)

  1. Read gyro + magnetometer, fuse them into a stable heading estimate (a complementary filter is enough — gyro for short-term smoothness, magnetometer to correct long-term drift).
  2. Master: broadcast its own heading + desired speed over UDP.
  3. Follower: receive target heading, compute error = target − own heading, run a small PID (a proportional term alone often works to start) to set differential motor speeds.
  4. Loop at ~20–50 Hz.

Because every robot computes its own heading from its own sensors, small calibration differences between robots just show up as slightly different turning delays — a useful teaching point about sensor variance.

5. Calibration

  • Magnetometers need per-robot hard/soft-iron calibration (spin the robot slowly in place, record min/max readings, compute an offset). Skipping this is the most common reason heading-following demos fail.
  • Mount IMUs away from motors — motor magnets distort magnetometer readings, a subtle bug students will hit if the IMU sits too close to a DC motor.

6. Suggested Build Order

  1. Get one robot reading a stable fused heading (calibrate mag, tune complementary filter) — a good first milestone on its own.
  2. Get master → single follower UDP link working with a fixed test heading.
  3. Add live master heading broadcast + follower PID steering.
  4. Scale to 3+ followers (UDP broadcast makes this free — no code changes needed).
  5. Extension ideas: forward speed matching, obstacle avoidance, loose formation offsets.

7. IMU Comparison

The class solders their own sensor boards regularly, so solderless connectors are not a deciding factor here — cost and library support matter more.

Sensor Price (each) Sensor fusion Connector MicroPython support Notes
ICM-20948 $14.95 (Adafruit) / ~$18–20 (SparkFun) Raw data + onboard DMP (optional) STEMMA QT / Qwiic, solderless Good — several community libraries Successor to the discontinued MPU-9250; accel/gyro + AK09916 magnetometer in one package. Best balance of price and teachability.
BNO085 $19.95 (Adafruit) Built-in — outputs orientation directly STEMMA QT / Qwiic, solderless Thin — mostly Arduino/CircuitPython Fastest path to a working demo since sensor fusion is done onboard.
BNO055 $34.95 (Adafruit) Built-in STEMMA QT / Qwiic, solderless Thin — mostly Arduino/CircuitPython Same idea as BNO085 at a higher price — no real upside over it.
LSM9DS1 ~$19.95 (Adafruit) Raw data only STEMMA QT / Qwiic, solderless Reasonable Solid raw-data option if students are building their own filter.
L3GD20 + LSM303D ~$4 (clone boards) Raw data only Pin header, solder required Community libraries for each chip separately Two separate chips (gyro + accel/mag), so two libraries instead of one. At this price, a strong pick for outfitting a full class of robots.
MPU-9250 / GY-91 clone ~$6–10 (generic) Raw data only Pin header, solder required Widely used, many tutorials Cheap, but inconsistent QC and a known-finicky magnetometer (AK8963).

Recommendation

Given the class already solders sensor boards, the L3GD20 + LSM303D is a strong choice: roughly a quarter of the price of the STEMMA QT options, with comparable sensing capability for this project. The main added cost is software — two separate I2C devices and two libraries to fuse into one heading estimate, which doubles as a useful lesson in combining multi-chip sensor data.

If the priority shifts toward getting a demo running fastest (e.g., a single student prototype under time pressure), the BNO085 ($19.95) remains the quickest path, since it outputs a ready-to-use heading with no filter-writing required.