Semi-autonomous Teleoperation using Differential Flatness of a Crane Robot for Aircraft In-Wing Inspection

TL;DR

The paper tackles ergonomic and safety challenges in aircraft-wing inspection by introducing a crane robot that traverses the entire wing through a stringer channel, enabling teleoperation from outside the confined space. It leverages differential flatness to generate reduced-oscillation, collision-free camera-payload trajectories and combines feedforward flatness-based inputs with limited state feedback in a semi-autonomous controller. Key contributions include a rigorous flatness-based trajectory generation framework, a collision-avoidance mechanism using obstacle bounding boxes, and real-time trajectory planning that eliminates collisions and reduces oscillations, leading to measurable gains in inspection efficiency. The approach is validated through autonomous swing-compensation experiments (89% oscillation reduction) and user trials with 12 participants showing zero collisions and a 18.7% efficiency improvement when neglecting collisions, highlighting practical impact for safe and efficient confined-space inspections.

Abstract

Visual inspection of confined spaces such as aircraft wings is ergonomically challenging for human mechanics. This work presents a novel crane robot that can travel the entire span of the aircraft wing, enabling mechanics to perform inspection from outside of the confined space. However, teleoperation of the crane robot can still be a challenge due to the need to avoid obstacles in the workspace and potential oscillations of the camera payload. The main contribution of this work is to exploit the differential flatness of the crane-robot dynamics for designing reduced-oscillation, collision-free time trajectories of the camera payload for use in teleoperation. Autonomous experiments verify the efficacy of removing undesired oscillations by 89%. Furthermore, teleoperation experiments demonstrate that the controller eliminated collisions (from 33% to 0%) when 12 participants performed an inspection task with the use of proposed trajectory selection when compared to the case without it. Moreover, even discounting the failures due to collisions, the proposed approach improved task efficiency by 18.7% when compared to the case without it.

Figures (10)

• Figure 1: The crane robot performs inspection along the entire wing by traversing in a single stringer channel. (Top-left) Partial view of an aircraft wing bay with crane robot moving inside the channel between two stringers, suspending a camera payload for inspection. The wing bays are partitioned into separate spaces by ribs and each space is typically accessed through narrow access holes. (Center) A two-dimensional schematic of the wing with a sample stringer channel highlighted in yellow, spanning the length of the wing, which provides access (over the ribs, with the camera stowed) between adjacent bays. (Bottom left, Section A-A) Cross section with the stowed crane robot in the stringer channel (highlighted in yellow) formed below the upper skin and between adjacent stringers enabling passage over the ribs. (Bottom right, Section B-B) Crane robot in the stringer channel (yellow highlight).
• Figure 2: Installation of the crane robot in the stinger channel. (a) The crane robot frame segmented in two pieces with widths $w_1=3.75$ in and $w_2=2.75$ in and thickness less than $2.5$ in so each segment can be installed diagonally through the channel opening of width $w_c=3.5$ in. (b) The installation is completed by connecting the segmented frame with a threaded rod (red). The width of the connected segments is larger than the channel opening width, $w_c$, enabling the crane robot to drive on the stringer flanges with its vertical wheels (cyan) while its side wheels (blue) contact the stringer web to correct channel misalignment.
• Figure 3: Crane-robot's control scheme. The external camera at the access hole measures the cart position, $x$, the payload length, $l$, and the swing angle, $\theta$, from the fiducial markers at a sampling rate of 33 Hz, which intermittently updates feedback collected from the motor encoders at a sampling rate of 1.3 kHz to construct the crane robot's states, $X$. From the operator workstation, the feedforward input, $F_{ff}$, and desired states, $X_d$, are specified through a joystick interface by the operator observing the external camera and camera payload image. The combined feedforward force, $F_{ff}$, and feedback force, $F_{fb}$, is the applied crane-robot input, $F$.
• Figure 4: Fiducial marker configuration for global feedback. Crane robot states and outputs are computed using a combination of marker coordinate measurements and crane-robot kinematics.
• Figure 5: Comparison of residual oscillations for slow and fast trajectories, without swing-dynamics compensation. (a) The crane-robot schematic. The system inputs are the force on the cart, $f_1$, and the force on the payload, $f_2$, and the system outputs are the horizontal and vertical positions of the payload, $y_1$ and $y_2$, respectively. The states of the system are the cart's position, $x$, the payload length, $l$, and the payload swing angle, $\theta$, along with their time derivatives. (b) The experimental confined space with snapshots of the crane robot prototype traversing the ramp-like trajectory which transitions the camera payload over a pipeline obstacle near an inspection location. (c) Comparative time responses of tracking the decoupled commands for the horizontal camera coordinate, $y_1$, and the vertical camera coordinate, $y_2$, at two different transition times, $T_t=4$ and $T_t=40$ seconds, plotted against normalized time, $\tau_n = \frac{\tau}{4T_t}$. (d) The residual oscillations upon arriving at the inspection point for the two transition times.

Theorems & Definitions (15)

• Remark 1
• Lemma 1
• proof
• Remark 2
• Corollary 1
• proof
• Remark 3
• Corollary 2
• proof
• Remark 4