ArduSoar: An Open-Source Thermalling Controller for Resource-Constrained Autopilots

Samuel Tabor, Iain Guilliard, Andrey Kolobov

TL;DR

ArduSoar is the first open-source soaring controller integrated into a major autopilot suite (ArduPilot), allowing small UAVs to autonomously detect and exploit thermals. By using a computationally efficient Extended Kalman Filter (EKF) to model rising air, it enables resource-constrained hardware to significantly extend flight times by "hitching a ride" on atmospheric energy.

What Problem Does This Paper Address?

Fixed-wing small Unmanned Aerial Vehicles (sUAVs) are strictly limited by their battery life or fuel capacity. While birds and glider pilots have long used thermals—rising columns of warm air—to stay aloft for hours without flapping or engines, bringing this capability to autonomous drones is difficult.

The challenges are twofold:

  1. Sensing the Invisible: Thermals are invisible and chaotic. A drone must distinguish between gaining altitude because it hit a thermal versus gaining altitude because the pilot pulled up on the elevator.
  2. Hardware Constraints: Most autonomous soaring research relies on heavy-duty sensors or powerful off-board computers. Most sUAVs run on "resource-constrained" flight controllers (like the Pixhawk) that have very limited processing power.

Before ArduSoar, there was no publicly available, integrated solution that allowed a standard hobbyist or commercial drone to perform these maneuvers out of the box.

Methods

The authors developed ArduSoar, which is now part of the ArduPlane firmware. Its operation follows a four-stage cycle: detection, identification, exploitation, and exit.

  • The Netto Variometer: To "see" the air, the system calculates the aircraft's total specific energy (potential + kinetic). By subtracting the aircraft's known "sink rate" (how fast it naturally falls at a given speed and bank angle, known as the polar curve), the controller can isolate the vertical movement of the air itself.
  • Simplified EKF: To identify the thermal's center and strength, ArduSoar uses a 4-variable Extended Kalman Filter. It estimates the thermal's strength, radius, and its North/East distance from the plane. Crucially, the authors simplified the math (linearizing the model) so it can run at 5Hz on low-power microcontrollers.
  • Drift Correction: The system accounts for wind, assuming the thermal drifts at the same velocity as the surrounding air mass, allowing the drone to "stay in the bubble" as it moves across the ground.
  • Mode Integration: ArduSoar toggles between AUTO (following waypoints) and LOITER (circling the thermal). It triggers the motor only when the altitude drops below a safety threshold (SOAR_ALT_MIN).

Datasets

The paper does not use a pre-existing static dataset. Instead, it validates the system through:

  • High-fidelity Simulations: Using Rayleigh-Bénard convection models to visualize how chaotic real thermals are compared to idealized "columns."
  • Real-world Flight Tests: Conducted between October 2017 and January 2018 in the US Pacific Northwest using a Parkzone Radian Pro sailplane equipped with a Pixhawk flight controller. The authors performed dozens of flights, successfully "catching" thermals in 21 of them.

Key Insights and Ideas

  • Robustness over Complexity: Real thermals are messy and irregular, but the authors found that a simple "bell-shaped" mathematical model is "good enough" for an EKF to track and exploit them effectively.
  • The Fixed Radius Strategy: Theoretically, a drone should change its turn radius to find the strongest lift. However, the authors discovered that on resource-constrained hardware, the EKF's estimate of thermal radius becomes noisy once the drone is centered. Their "aha moment" was realizing that using a fixed, small loiter radius (15m for their testbed) is more robust and easier to tune than trying to optimize the radius dynamically.
  • Passive vs. Active: ArduSoar is a "reactive" controller. It doesn't actively hunt for thermals; it simply monitors the air while performing its normal mission and "pounces" when it happens to fly through lift.

Key Results

  • Successful Altitude Gains: The flight tests showed the drone successfully entering LOITER mode upon hitting lift, spiraling upward (forming a "helix" path), and gaining significant altitude without motor power.
  • Versatility: The controller proved robust even in non-ideal conditions, such as orographic lift (wind hitting a hill and going up), even though the underlying model was designed for thermal columns.
  • Efficiency: In one highlighted flight (Figure 6), the drone maintained or gained altitude for over 30 minutes with the motor off for the vast majority of the time, demonstrating the potential for massive range extension.

Limitations and Future Work

  • No Active Exploration: Currently, if the drone is 50 meters away from a massive thermal, it won't know it's there. Future work could involve "deliberate exploration" strategies where the drone deviates from its path to scout likely thermal locations.
  • Manual Tuning: Several parameters (like the polar curve and the fixed loiter radius) still require manual calibration for different airframes.
  • Static vs. Dynamic: This paper focuses on static soaring (rising air). It does not address dynamic soaring (extracting energy from wind gradients), which is a much more complex control problem.

Reference: Tabor, S., Guilliard, I., & Kolobov, A. (2018). ArduSoar: an Open-Source Thermalling Controller for Resource-Constrained Autopilots. arXiv:1802.08215v2 [cs.RO].

Summary updated

Abstract

Autonomous soaring capability has the potential to significantly increase time aloft for fixed-wing UAVs. In this paper, we introduce ArduSoar, the first soaring controller integrated into a major autopilot software suite for small UAVs. We describe ArduSoar from the algorithmic standpoint, outline its integration with the ArduPlane autopilot, discuss parameter tuning for it, and conduct a series of flight tests on real sUAVs that show ArduSoar's robustness even in highly nonideal atmospheric conditions.
BibTeX
@inproceedings{tabor2018ardusoar,
  author = {Samuel Tabor and Iain Guilliard and Andrey Kolobov},
  title = {{ArduSoar: An Open-Source Thermalling Controller for Resource-Constrained Autopilots}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2018},
  url = {https://openreview.net/forum?id=dQkUidV4P6V},
}

Loading related content...

Figure