Onboard Safety Guarantees for Racing Drones: High-Speed Geofencing With Control Barrier Functions

Andrew Singletary, Aiden Swann, Yuxiao Chen, Aaron D. Ames

TL;DR

This paper introduces a novel safety filter for racing drones that uses Control Barrier Functions (CBFs) to guarantee that the drone stays within a defined geofenced area, even at speeds over 100 km/h. The key innovation is an "implicitly defined CBF" that leverages a backup controller, allowing for gradient-free safety checks on a low-power microcontroller, making it practical for real-world drone racing.

What Problem Does This Paper Address?

Drone racing is a rapidly growing sport, but it's currently dominated by human pilots. While autonomous racing drones are advancing, ensuring safety for human-piloted drones, especially at high speeds and during aggressive maneuvers, remains a significant challenge. Existing safety methods often rely on complex planning or computationally intensive algorithms that are not feasible on the small, low-power microcontrollers found on racing drones. This paper aims to bridge this gap by providing robust, onboard safety guarantees that allow pilots to fly aggressively without fear of crashing, effectively creating a "safety net" that intervenes only when necessary.

Methods

The core of the proposed solution is a Control Barrier Function (CBF) framework, which mathematically guarantees that a system's state remains within a predefined "safe set." However, traditional CBFs can be computationally demanding, requiring gradient calculations and optimization solvers. To overcome this, the authors introduce an implicitly defined CBF that leverages a backup controller.

Here's how it works:

  • Implicit CBF: Instead of directly calculating gradients, the safety condition is evaluated by simulating the drone's dynamics under a pre-defined "backup controller" for a short duration. This backup controller is designed to bring the drone back to a safe state or a safe region.
  • Backup Controller: This controller is designed to be simple and computationally cheap. It's activated when the drone gets close to the boundary of the safe set. Crucially, it's designed to move the drone away from the boundary, not just stop it there, preventing the pilot from getting "stuck."
  • Smooth Safety Filtering: The system doesn't abruptly switch to the backup controller. Instead, a smooth mixing function, parameterized by a decay rate $\beta$, gradually blends the pilot's desired input with the backup controller's input as the drone approaches the boundary. This ensures smooth, non-jarring interventions.
  • High-Speed Modification: For very high speeds, a modification is introduced to the mixing function. It scales the safety condition by the inverse of the velocity towards the barrier, preventing the filter from becoming overly conservative at lower speeds while still ensuring safety at high speeds.

This approach avoids the need for gradient computations and optimization solvers, making it suitable for low-power onboard hardware.

Datasets

Not discussed in the paper. The evaluation relies on simulation and real-world flight tests.

Key Insights and Ideas

  • Practical Safety on Microcontrollers: The most significant insight is demonstrating that formal safety guarantees (CBFs) can be implemented on extremely resource-constrained hardware (a 10g microcontroller) for high-speed drone flight. This opens up possibilities for safety-critical applications on small aerial vehicles.
  • Gradient-Free Safety: The implicit CBF approach, by avoiding gradient calculations, dramatically reduces computational load. This is a crucial step towards making advanced safety techniques accessible for real-time applications on embedded systems.
  • Smooth, Non-Intrusive Intervention: The smooth mixing function ensures that the safety filter only subtly alters the pilot's commands when necessary. This preserves the pilot's control authority and makes the system feel natural to operate, rather than being overly restrictive.
  • Backup Controller Design Matters: The design of the backup controller is critical. It needs to be simple, effective at returning the drone to safety, and importantly, it must actively move the drone away from the boundary to avoid getting stuck.

Key Results

  • Guaranteed Geofencing at High Speeds: The system successfully enforced geofencing on a 7" racing drone at speeds exceeding 100 km/h (specifically, 104 km/h in one test).
  • Robustness to Free Fall: The safety filter was also effective in recovering from a 70m free fall, bringing the drone to a safe hover before impact.
  • Minimal Pilot Intervention: The filter only intervenes when the drone approaches the boundary, and the interventions are smooth, allowing pilots to push their limits. In tests, the $\lambda$ parameter (which indicates the degree of filtering) never reached zero, meaning the pilot always retained some control.
  • Computational Efficiency: The filter runs in approximately 350 $\mu$s on the Teensy 4.1 microcontroller, operating at a 400 Hz update rate. This is orders of magnitude faster than traditional CBF implementations requiring optimization.
  • Real-World Validation: Extensive flight tests demonstrated the reliability and consistency of the method, with safety maintained across multiple barrier engagements in each flight.

Limitations and Future Work

  • Known Environments: The current method relies on a predefined safe set (geofence). It does not handle dynamic obstacles or unknown environments.
  • Limited Pilot Authority Near Boundary: While the filter is designed to be minimally conservative, the $\lambda$ parameter can become low near the boundary, meaning the pilot's input is significantly modified. The authors note that this is not ideal for collision avoidance in dynamic environments.
  • Future Work: The authors propose extending the system to dynamically construct safe sets using onboard vision, enabling collision avoidance with dynamic obstacles. They also suggest exploring multi-robot collision avoidance for drone races. Additionally, they aim to investigate how the ability of traditional CBFs to utilize gradient information for faster motion along boundaries could be incorporated into their framework for improved performance in dynamic scenarios.

Summary updated

Abstract

This letter details the theory and implementation behind practically ensuring safety of remotely piloted racing drones. We demonstrate robust and practical safety guarantees on a 7” racing drone at speeds exceeding 100 km/h, utilizing only online computations on a 10 g micro-controller. To achieve this goal, we utilize the framework of control barrier functions (CBFs) which give guaranteed safety encoded as forward set invariance. To make this methodology practically applicable, we present an implicitly defined CBF which leverages backup controllers to enable gradient-free evaluations that ensure safety. The method applied to hardware results in smooth, minimally conservative alterations of the pilots’ desired inputs, enabling them to push the limits of their drone without fear of crashing. Moreover, the method works in conjunction with the preexisting flight controller, resulting in unaltered flight when there are no nearby safety risks. Additional benefits include safety and stability of the drone when losing line-of-sight or in the event of radio failure.
BibTeX
@article{singletary2022onboard,
  author = {Andrew Singletary and Aiden Swann and Yuxiao Chen and Aaron D. Ames},
  title = {{Onboard Safety Guarantees for Racing Drones: High-Speed Geofencing With Control Barrier Functions}},
  journal = {IEEE Robotics Autom. Lett.},
  year = {2022},
  url = {https://openreview.net/forum?id=nb2tG8--Vi},
}
Figure