Source-linked AI summary
Vision-Based Control of a Tether-Suspended Aerial Radiation Sensing Payload
Ian Snider, Brian J. Quiter, Emil Rofors, Mark W. Mueller
TL;DR
Radiation surveys need detectors close to the source, but low-flying propellers can disturb contaminated ground, while a tethered payload is only indirectly controllable. The paper presents vision-based payload-position control for a tethered detector beneath a heavy-lift drone. In outdoor wind tests, transit position RMSE fell by about 20% against a vehicle-referenced baseline, although peak error was higher at waypoint arrival.
Problem
Close source proximity improves radiation-detector sensitivity, but low flight risks resuspending contaminated particulates and tethered payloads are difficult to control indirectly.
Source: §I · p. 1Method
A downward-facing camera tracks a payload ring, a Kalman filter estimates payload motion, and an LQI controller regulates payload position along a pilot-commanded trajectory.
Source: §I · p. 2, C. Pilot Aid · p. 3, §IV · p. 4Results
About 20%: payload-aware control reduced transit position RMSE from 0.88 m to 0.70 m versus the vehicle-referenced baseline, but struggled with larger waypoint-transition swings.
Source: §V · p. 6, §VI · p. 7Takeaways & Limitations
Outdoor flight tests under wind showed reduced payload tracking error against the ArduPilot vehicle-referenced baseline, providing a baseline for more advanced model-predictive control.
Source: §VI · p. 7The approach assumes small swing angles; anticipated complex-terrain surveys require failsafes, and heavy-lift control advances must be validated carefully.
Source: §VI · p. 7
Abstract
from arXiv · showhide
Aerial radiation surveys achieve higher sensitivity when the radiation detector is held close to the ground. Detector sensitivity falls off roughly with the inverse square of the distance to the source, so a detector flown high is slower to reach a given minimum detectable activity. Flying the vehicle low puts the propellers near the ground, where downwash can disturb the surveyed area and resuspend contaminated particulates. Tether suspension decouples the detector from the vehicle altitude, but leaves the payload unactuated and only indirectly controllable. We therefore present a vision-based control approach for an aerial sensing payload suspended on a tether beneath a heavy-lift drone. Because a survey plan is decided as radiation detections arrive, we design a pilot aid for commanding the survey trajectory manually with a handheld transmitter. The controller regulates the payload, rather than the vehicle, onto that trajectory. The system uses onboard sensors with a downward-facing camera fixed to the drone body tracking a ring marker on the payload. A four-state Kalman filter estimates the tether swing angles and rates from payload bearing measurements, and a linear quadratic regulator with integral action takes the payload position as the regulated output. In outdoor flight tests under wind, the payload-aware controller reduced payload tracking error during transit by 20% when compared against a vehicle-referenced baseline, with the cost of higher peak error on arrival at a waypoint.
I. INTRODUCTION
Radiation surveys benefit from placing detectors close to the source, but low-flying propellers can resuspend contaminated soil. Tethered payload control is challenging, and this work addresses payload-position regulation and modeling for that setting.
I. INTRODUCTION
- Detector sensitivity is roughly proportional to inverse-square source distance, motivating low-altitude surveys, where rotor downwash can resuspend contaminated particulates.Tether suspension keeps the vehicle at a safer altitude relative to cluttered or contaminated terrain.Source: §I · p. 1
- The study presents an outer-loop controller that regulates payload position through acceleration commands to an unmodified ArduPilot inner loop, and reports outdoor flight tests comparing payload-aware control with a vehicle-referenced baseline.The effect of wind on outdoor heavy-lift tethered payload systems remains largely uncharacterized.Source: §I · p. 2
- Unlike cargo-transport work that treats the suspended payload as a disturbance, this study regulates payload position using onboard vision and evaluates outdoor performance under wind.The contributions include a payload-bearing estimator that recovers a ring center from partial arcs despite tether occlusion, and a pilot aid for commanding the payload trajectory.Source: §I · pp. 1–2
II. MODELING
- The model treats the drone and payload as point masses joined by a taut, massless tether, and restricts swing analysis to small angles.Linearization about hover yields decoupled pendulum equations; damping is omitted and aerodynamic drag is absorbed into estimator process noise.Source: §II · p. 2
III. CONTROLLER DESIGN
The controller regulates payload position to a desired setpoint while retaining the existing ArduPilot infrastructure for safety.
III. CONTROLLER DESIGN
- The outer-loop controller sends acceleration setpoints to the existing ArduPilot inner loop to regulate the payload at a desired position.The existing flight-control infrastructure is retained to limit vehicle behavior for safety.Source: §III · p. 2
A. ArduPilot Inner-Loop
The ArduPilot inner loop receives outer-loop acceleration commands while preserving a pilot-recoverable failsafe path.
A. ArduPilot Inner-Loop
- ArduPilot receives acceleration commands at 50 Hz in GUIDED mode, and the pilot can recover control by switching out of that mode.Position and velocity correction are disabled, the command is feedforward, and ArduPilot applies a 10 m/s^3 jerk limit.Source: A. ArduPilot Inner-Loop · p. 2
- The architecture separates payload estimation and acceleration commands at 50 Hz, ArduPilot inner-loop control, and a 48 fps vision thread.The system also receives external input from the reference generator, pilot, and wind.Source: Figure 3 · p. 3
B. Acceleration Controller
The acceleration controller uses integral LQI feedback to regulate payload position and address persistent disturbances, with conditional integration to limit wind-up.
B. Acceleration Controller
- The controller regulates payload position as the output of a state-space model combining vehicle double-integrator dynamics with linearized pendulum dynamics.The state includes vehicle position and velocity and payload state; the tether’s constant vertical offset is part of equilibrium, not the output.Source: B. Acceleration Controller · p. 3
- Integral action on payload position error addresses the steady-state offset that proportional state feedback leaves under persistent disturbances such as wind.The controller uses an LQI cost weighting payload position error, integrated error, and control effort.Source: B. Acceleration Controller · p. 3
- Conditional integration prevents integral wind-up when actuator saturation prevents commanded acceleration from being achieved.Integration is enabled only inside an error band, and the integral command is capped.Source: B. Acceleration Controller · p. 3
C. Pilot Aid
The pilot commands the payload reference trajectory through transmitter sticks, while the controller tracks position and velocity setpoints. Integral action can produce large swings, especially when the pilot cannot see the generated reference.
C. Pilot Aid
- Transmitter stick signals set a bounded payload velocity, which is ramped by an acceleration limit and integrated into a position setpoint.The controller reference is a position and velocity setpoint along the trajectory.Source: C. Pilot Aid · p. 3
- Integral action can contribute to large swings during manual control, so LQR may be preferable when the pilot cannot see the generated reference.The authors nevertheless use LQI to analyze reference-tracking capability.Source: C. Pilot Aid · p. 3
IV. PAYLOAD ESTIMATION
Payload-state estimation uses a downward-facing camera to measure a ring marker’s bearing and a Kalman filter to estimate the payload state from bearing and vehicle acceleration.
IV. PAYLOAD ESTIMATION
- The controller requires accurate payload-state information, estimated from a red ring marker on the payload periphery using a downward-facing camera.The camera provides bearing but not range.Source: §IV · p. 4
- A Kalman filter combines camera bearing measurements with measured vehicle acceleration, treated as a known input, to estimate payload state.Source: §IV · p. 4
A. Process Model
The process model represents payload pendulum motion with four swing states and uses vehicle acceleration as an input, while modeling disturbances as noise on swing rates.
A. Process Model
- The four-state process model comprises the payload’s swing angles and swing rates.Source: A. Process Model · p. 4
- Measured horizontal vehicle acceleration, transformed into the inertial frame, is the process-model input and keeps the filter state restricted to the payload.The state-space dynamics are written as ˙ξ = Fξ + Gu.Source: A. Process Model · p. 4
- Aerodynamic drag, tether flexing, and wind gusts are modeled as unmeasured accelerations entering only the swing-rate states.The rate-state noise intensity is 0.01 for both swing rates; swing states receive no process noise.Source: A. Process Model · p. 4
B. Measurement
The measurement model compares camera-frame payload direction with observed bearing to update estimates of the payload’s swing states.
B. Measurement
- The Kalman filter estimates payload swing states using payload direction as the measured quantity.Source: B. Measurement · p. 4
- Under the small-angle restriction, depth is approximated as one and perspective division is neglected, making the measurement model affine in the swing state.The affine offset is the camera-view projection of the local vertical and vanishes at hover.Source: B. Measurement · p. 4
- The predicted payload direction is rotated into the camera frame before comparison with the camera’s normalized-image-plane bearing.The camera measurement uses the horizontal direction components divided by depth.Source: B. Measurement · p. 4
- Measurement updates run only on frames with an available bearing observation, while prediction runs at every controller step.Measurement noise σx = σy = 0.01 was found to work well.Source: B. Measurement · p. 4
C. Image Processing
A downward-facing camera tracks a red ring on the payload, fitting its center from visible arc segments despite occlusion. The resulting pixel location is converted to a payload bearing; range is assumed fixed along that bearing.
C. Image Processing
- The ring center is recovered by fitting a circle to the largest visible arc after thresholding red pixels and closing small gaps.Connected-component contours provide the edge points for the fit.Source: C. Image Processing · pp. 4–5
- A 70° minimum arc-coverage gate and a tether-length-based radius check reject false detections before the camera-pivot offset and calibrated intrinsic matrix convert pixels to bearing.Source: C. Image Processing · p. 5
- The filter uses the first two components of the bearing measurement; range is not measured because the payload is assumed to lie a fixed distance L along the bearing.Source: C. Image Processing · p. 5
V. EXPERIMENT
In outdoor tests under wind, the payload-aware controller reduced transit tracking error relative to the vehicle-referenced ArduPilot baseline, but had higher peak error at waypoint arrival. Tests also showed reliable ring detection despite tether occlusion and demonstrated manual piloting with the payload-aware controller.
V. EXPERIMENT
- Four outdoor square-trajectory tests used 15 m legs, 15 s corner hovers, and a 1.25 m/s command; south winds averaged about 3 m/s with gusts to about 6 m/s.Two tests used each controller, and the tests alternated under gusty conditions.Source: §V · p. 5, Figure 5 · p. 5
- 98.89% of 21,172 frames yielded a valid ring fit at 48 fps despite continuous partial tether occlusion.The longest dropout was 0.44 s.Source: §V · p. 5
- At waypoint transitions, peak swing reached 12.97° with payload-aware control versus 12.74° with the baseline; the baseline had lower overall hover RMSE, while payload-aware control settled nearer the reference.The baseline’s momentum carried the payload over the waypoint, while payload-aware control acted against payload motion.Source: §V · pp. 5–6, Table II · p. 6
- 0.72 m to 0.66 m mean per-leg tracking RMSE and 0.10 m to 0.05 m standard deviation favored payload-aware control, though mean differences were comparable to variation between legs.The worst leg was 0.72 m for payload-aware control versus 0.86 m for the baseline.Source: §V · p. 6
- 0.88 m to 0.70 m (about 20%): payload-aware control reduced transit position RMSE versus the vehicle-referenced ArduPilot baseline.The transit RMSE was just outside the baseline’s standard deviation.Source: §V · p. 6, Table II · p. 6
- In a short piloting-aid loop, slow velocity ramps let the pilot resume motion before the payload settled; the pilot reported payload-aware control was easier to fly.The authors note that reference generation could be improved with smooth S-curves.Source: §V · p. 6, Figure 7 · p. 6
VI. CONCLUSION AND FUTURE WORK
The work combines an LQI acceleration controller with vision-based payload-bearing estimation. Outdoor tests reduced payload tracking error versus a vehicle-referenced baseline, while larger swings remained a challenge at waypoint transitions.
VI. CONCLUSION AND FUTURE WORK
- The system uses an LQI acceleration controller to regulate tethered-payload position and vision-based bearing estimation with a Kalman filter that sustains estimates during brief dropouts.Source: §VI · pp. 6–7
- Outdoor flight tests under wind reduced payload tracking error versus the ArduPilot vehicle-referenced baseline, though larger swings challenged the controller at waypoint transitions.The conclusion identifies the LQI acceleration controller and vision-based payload-bearing estimation as contributions.Source: §VI · p. 7
- A real-world radioactive contamination survey remains future work, with slow, methodical maneuvers expected to keep swing angles within the small-angle assumption.The system is a baseline for more advanced model-predictive control; the authors say heavy-lift control advances require careful validation.Source: §VI · p. 7