💻 computer science

Airflow Source Seeking on Small Quadrotors Using a Single Flow Sensor

This paper presents a source-seeking system for sub-100g quadrotors that utilizes a custom airflow sensor to implement a modified "Cast and Surge" algorithm, enabling reliable detection and navigation toward flow sources to complement traditional chemical plume tracking.

Original authors: Lenworth Thomas, Tjaden Bridges, Sarah Bergbreiter

Published 2026-01-23

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Lenworth Thomas, Tjaden Bridges, Sarah Bergbreiter

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a tiny, palm-sized drone (weighing less than a bag of sugar) trying to find the source of a smell or a gas leak in a room. Usually, these drones rely on "chemical noses" (gas sensors) to sniff out the danger. But on such a small drone, these chemical noses are slow and often confused by the wind blowing from the drone's own spinning propellers. It's like trying to smell a flower while standing in front of a leaf blower.

This paper presents a clever workaround: instead of just waiting for the smell, the drone learns to feel the wind to find the source.

Here is how they did it, explained simply:

1. The "Feather" Sensor

The researchers built a custom sensor that looks a bit like a tiny, flexible feather mounted on top of the drone.

How it works: When wind hits this "feather," it bends and twists. Inside the base, there is a magnet and a sensor that measures exactly how much the feather moved and which way it turned.
The Magic: This gives the drone two pieces of information instantly: how strong the wind is and which direction it is coming from. Unlike chemical sensors that take time to react, this "feather" reacts immediately to the air moving around it.

2. The "Cast and Surge" Strategy (The Moth Method)

The drone uses a navigation strategy inspired by nature, specifically how moths find mates by following scent trails. The researchers call their version "Vector Surge."

Think of the drone's search pattern like a person looking for a campfire in the dark:

The "Cast" (Searching): If the drone isn't feeling any wind, it starts "casting." It moves left, then forward, then right, widening its search path like a fisherman casting a net. It keeps doing this until it feels a breeze.
The "Surge" (Chasing): The moment the "feather" sensor feels the wind, the drone stops wandering. It instantly turns its nose toward the wind (like a weather vane) and flies straight upwind, hoping to find the source.
The Safety Net: If the drone loses the wind while flying toward the source (maybe it flew out of the breeze), it immediately stops, turns around, and goes back to "casting" until it finds the wind again.

3. What They Proved

The team tested this on a small DJI Tello drone in a large indoor room with motion-tracking cameras to see exactly where the drone went.

Sensitivity: They proved the drone could feel very gentle breezes (as light as a calm day) even while hovering.
Speed: When the wind started, the drone could turn itself to face the wind in about 10 seconds. This is fast enough to catch a typical gust of wind before it blows away.
Success: In multiple trials where the drone started in random spots and facing random directions, it successfully found the wind source and flew toward it every time.

The Bottom Line

This research shows that by giving a tiny drone a simple "feather" to feel the wind, it can navigate much smarter than if it relied only on slow chemical sensors. It's a foundational step toward future drones that can use this wind-sensing ability to help chemical sensors work faster and more accurately, allowing them to find gas leaks or pollution sources quickly and safely around humans.

Limitations noted by the authors:

The current sensor only works well for wind blowing from the side (2D). It can't perfectly tell if wind is coming from directly above or below yet.
The drone doesn't have obstacle avoidance built-in yet, so in a cluttered room, it might bump into things if it doesn't find the wind quickly.
This specific test was done with wind, not actual chemicals, though the goal is to combine this with chemical sensors later.

Technical Summary: Airflow Source Seeking on Small Quadrotors Using a Single Flow Sensor

Problem Statement
As environmental disasters and hazardous emissions become more frequent, the ability to locate the sources of pollutants or harmful particulates via plume tracking is critical. While Unmanned Aerial Vehicles (UAVs) offer advantages in 3D space coverage and safety near humans, small quadrotors (< 100 g) face significant challenges in source-seeking. Traditional approaches relying on chemical gas sensors suffer from slow response times, poor sensitivity, and susceptibility to rotor-induced flow interference. Furthermore, many existing airflow sensing methods are designed for larger UAVs or require complex sensor arrays, making them unsuitable for small, palm-sized platforms operating in confined spaces.

Methodology
This work presents a system integrating a custom, lightweight flow sensor with a small quadrotor (DJI Tello, 80 g) to enable airflow source-seeking without relying solely on chemical gradients.

Sensor Design and Integration: The system utilizes a custom flow sensor inspired by prior work [21], featuring a fin with a magnet suspended on a planar spring. A magnetic field sensor measures the fin's displacement and rotation in three axes, providing estimates of both airflow magnitude and direction. The sensor and a 3D-printed modular mount weigh 15.6 g, keeping the total payload under 100 g. The sensor is mounted above the quadrotor's center to minimize interference from rotor downwash.
Signal Processing: A microcontroller (Seeed Studio XIAO nRF52840) samples magnetic field components ( $B_x$ and $B_y$ ) at 40 Hz. After a takeoff calibration to zero offsets, a 10-point moving average filter reduces noise. The airflow angle ( $\theta$ ) relative to the quadrotor is calculated using the inverse tangent of the ratio of the filtered $B_y$ and $B_x$ components, adjusted by 180° to determine the upwind direction.
Control Algorithm (Vector Surge): The authors implemented a bio-inspired "Vector Surge" algorithm, a modification of the "Cast and Surge" approach. The algorithm operates in three states:
1. Cast: If no flow is detected, the quadrotor performs translational movements (left, forward, right) with increasing duration to widen the search area.
2. Reorient: Upon detecting flow above a magnitude threshold (4.6 mT), the system uses a PD controller to rotate the quadrotor's yaw until the airflow aligns with the front of the vehicle.
3. Surge: Once oriented, the quadrotor moves upwind. If the flow signal is lost or a maximum magnitude is reached, the system returns to the Cast state.

Key Contributions
The primary contribution of this work is the successful integration of a single, directional flow sensor with a bio-inspired source-seeking algorithm on a sub-100 g quadrotor. Unlike previous methods that rely on slow chemical sensors or large sensor arrays, this approach uses direct airflow measurement to rapidly detect and navigate toward a source. The system provides both magnitude and direction data, enabling "flow servoing" to orient other potential sensors (e.g., gas sensors) toward the source, thereby potentially enhancing their sensitivity and response time.

Experimental Results
The system was evaluated in an indoor motion capture arena with randomized starting positions and orientations relative to a fan.

Sensor Characterization: In-flight tests confirmed the sensor could reliably estimate the angle of incident airflow across velocities ranging from 0.2 m/s to 1.24 m/s (Beaufort Scale 0–1). While raw data was noisy, the moving average filter provided a stable estimate that closely tracked the quadrotor's yaw rotation.
Reorientation Performance: The quadrotor successfully reoriented toward the airflow source from various initial angles (45°, 90°, 180°). In most trials, the system reduced the orientation error to within 20° in approximately 10 seconds, a timeframe comparable to half the duration of a typical sustained gust.
Source-Seeking Trials: Ten trials of the Vector Surge algorithm demonstrated the system's ability to find a flow source from random starting locations. The quadrotor successfully detected flow, reoriented, and surged toward the source in 38 to 55 seconds. In cases where the quadrotor drifted out of the flow during the surge, the algorithm successfully returned to the "Cast" state to reacquire the flow.

Significance and Limitations
The authors claim this work provides a foundation for future platforms that can combine flow sensing with other environmental sensors to enable richer plume tracking data collection. The ability to rapidly reorient a platform based on airflow direction addresses the latency issues inherent in chemical sensing.

However, the paper modestly notes several limitations:

2D Constraint: The current sensor design provides 2D flow vectors; 3D source-seeking remains future work.
Flow Separation: The system currently struggles to distinguish between flow from an external source and flow induced by the quadrotor's own motion or rotors, particularly in complex environments.
Obstacle Avoidance: The current implementation lacks obstacle avoidance, which led to manual termination of trials when the quadrotor approached arena walls.
Chemical Integration: While the system aids flow detection, the authors note that using this approach with chemical sensing would currently limit source-seeking to scenarios where a chemical of interest is present, suggesting the algorithm is most immediately suitable for outdoor environments where flow sources are distinct.

1. The "Feather" Sensor

2. The "Cast and Surge" Strategy (The Moth Method)

3. What They Proved

The Bottom Line

More like this