From Bench to Flight: Translating Drone Impact Tests into Operational Safety Limits

Indoor micro-aerial vehicles (MAVs) are increasingly used for tasks that require close proximity to people, yet practitioners lack practical methods to tune motion limits based on measured impact risk. We present an end-to-end, open toolchain that converts benchtop impact tests into deployable safety governors for drones. First, we describe a compact and replicable impact rig and protocol for capturing force-time profiles across drone classes and contact surfaces. Second, we provide data-driven models that map pre-impact speed to impulse and contact duration, enabling direct computation of speed bounds for a target force limit. Third, we release scripts and a ROS2 node that enforce these bounds online and log compliance, with support for facility-specific policies. We validate the workflow on multiple commercial off-the-shelf quadrotors and representative indoor assets, demonstrating that the derived governors preserve task throughput while meeting force constraints specified by safety stakeholders. Our contribution is a practical bridge from measured impacts to runtime limits, with shareable datasets, code, and a repeatable process that teams can adopt to certify indoor MAV operations near humans.

💡 Research Summary

The paper addresses the pressing need for practical safety limits for indoor micro‑aerial vehicles (MAVs) that operate in close proximity to people. While standards such as JARUS, Transport Canada, and EASA require collision‑risk mitigation, they provide no concrete data or models for specific drone designs. To fill this gap, the authors present an end‑to‑end, open‑source toolchain that converts bench‑top impact measurements into runtime speed governors for drones.

First, a compact, reproducible impact rig is described. The rig uses an electric catapult to launch a drone‑holding trolley at controlled speeds (3–4 m/s). Just before impact a solenoid releases the drone onto a secondary rail, allowing free rebound. Sensors include three PCB Piezotronics load cells (triangular arrangement), a single‑axis accelerometer, a TFmini‑S velocity sensor, and a high‑speed camera (1000 fps). All signals are synchronized with a 5 V TTL trigger and sampled at up to 6.25 kHz, providing high‑resolution force‑time histories.

The experimental campaign covers twenty specimens: four DJI Avata, four Flywoo FlyLens, eight carbon‑rod Cognifly, and four bamboo‑rod Cognifly drones. Material properties (Young’s modulus, tensile strength, density, toughness) are tabulated, and each drone is tested in frontal impact and, for the carbon Cognifly, a 45° oblique impact. Data processing follows a four‑stage pipeline: (1) synchronization, (2) median‑plus‑Kalman filtering for velocity, (3) force threshold and video validation for impact detection, and (4) conservative impulse integration that discards negative force artefacts. Contact duration is cross‑checked with video zero‑crossings to ensure timing accuracy.

Because the sample size per configuration is small (n = 4), the authors fit second‑ or third‑order polynomial regressions to three key relationships: impact speed → peak force, speed → impulse, and speed → contact duration. Model quality is assessed with both R² and mean absolute error (MAE), the latter providing an absolute error bound crucial for safety‑critical applications. The resulting functions give an explicit mapping F(v) that predicts average impact force for any approach speed within the tested range.

Using these regressions, a ROS 2 “Safety Governor” node is implemented. The node subscribes to odometry and range data to estimate the current speed and whether a human is inside a collaborative zone. The collaborative zone radius follows the ISO TS 15066‑style equation S = v T_q + 3/2 v²/a + C, where T_q is perception‑to‑actuation latency, a is worst‑case deceleration, and C is a safety margin. Simultaneously, the impact‑based speed limit v_max is computed by solving the scalar equation m v (1 + ê(v)) b Δt(v) = F* for a user‑defined force threshold F* (e.g., 150 N for neck). The governor publishes the minimum of the ISO‑based and impact‑based limits on the /cmd_vel_limited topic, providing a last‑line safeguard even if perception fails.

Experimental results show clear hierarchies in peak forces: the bamboo‑TPU Cognifly achieves the lowest peak (≈84 N), well below ISO limits for neck (150 N) and chest (140 N). The safety governor, when activated, reduces the commanded speed only when necessary; overall mission throughput remains above 85 % of the unrestricted case, demonstrating that safety does not come at the cost of performance.

Limitations are acknowledged: the impact speed range is limited to 3–4 m/s, the rebound zone is only 300 mm, and the dataset is small, restricting extrapolation to higher‑speed or larger drones. Future work is suggested to extend the rig for higher speeds, incorporate more impact angles and surface materials, and explore nonlinear machine‑learning models for better prediction.

In summary, the paper delivers the first complete pipeline—from bench‑top impact testing to a deployable ROS 2 safety governor—enabling teams to certify indoor MAV operations with quantifiable, data‑driven speed limits. All datasets, scripts, and the ROS 2 node are released publicly, fostering reproducibility and further research in human‑drone collaborative safety.