PaperlayerView exact revision on arXiv

Source-linked AI summary

An Omnidirectional Aerial Manipulation Platform for Contact-Based Inspection

Karen Bodie, Maximilian Brunner, Michael Pantic, Stefan Walser, Patrick Pfändler, Ueli Angst, Roland Siegwart, Juan Nieto

arXiv:1905.03502v2cs.RO

TL;DR

Contact-based non-destructive inspection of concrete remains labor-intensive, motivating aerial systems that can interact with structures. This paper presents a fully actuated omnidirectional tilt-rotor platform with selective-inertia impedance control, and reports contact traversal using on-board surface estimation as well as validation for non-destructive testing.

  • Problem

    Contact-based inspection of aging concrete requires substantial human labor, road closures, and large support equipment, while aerial vehicles have mainly served visual inspection.

    Source: §I · p. 1

  • Method

    The paper combines a fully actuated omnidirectional tilt-rotor with a rigid end-effector and a 6 DoF impedance controller using selective apparent inertia.

    Source: §I · p. 1

  • Results

    The system traversed 0.53 m during surface contact at an average velocity of 0.17 m s−1, and experiments demonstrate its use for contact-based inspection.

    Source: D. Depth Servoing for Contact with… · p. 7, §VI · p. 8

  • Takeaways & Limitations

    Experiments validate the platform as a tool for contact-based non-destructive testing of concrete infrastructure.

    Source: §I · p. 1, §VI · p. 8

Abstract

from arXiv · show

This paper presents an omnidirectional aerial manipulation platform for robust and responsive interaction with unstructured environments, toward the goal of contact-based inspection. The fully actuated tilt-rotor aerial system is equipped with a rigidly mounted end-effector, and is able to exert a 6 degree of freedom force and torque, decoupling the system's translational and rotational dynamics, and enabling precise interaction with the environment while maintaining stability. An impedance controller with selective apparent inertia is formulated to permit compliance in certain degrees of freedom while achieving precise trajectory tracking and disturbance rejection in others. Experiments demonstrate disturbance rejection, push-and-slide interaction, and on-board state estimation with depth servoing to interact with local surfaces. The system is also validated as a tool for contact-based non-destructive testing of concrete infrastructure.

I. INTRODUCTION

Aging concrete infrastructure needs more inspection capacity, while contact-based non-destructive testing still involves substantial labor and support requirements. The paper presents an omnidirectional aerial manipulator and selective-inertia impedance control for contact inspection.

I. INTRODUCTION

  • Contact-based non-destructive testing remains labor-intensive and requires road closures and large supporting equipment, despite aerial vehicles already serving visual infrastructure inspection.The paper frames aging concrete and limited inspection capacity as motivating concerns.Source: §I · p. 1
  • An actively tilting hexarotor with a rigidly attached end-effector provides omnidirectional force and torque while retaining efficient horizontal hover.The system is designed for on-board computation, state estimation, power, and sensing for field deployment.Source: §I · p. 1
  • The proposed 6 DoF impedance controller uses selective apparent inertia to combine compliance in some directions with disturbance rejection in others.Source: §I · p. 1
  • Experiments validate precise interaction control and demonstrate the platform as a tool for contact-based non-destructive testing of concrete infrastructure.Source: §I · p. 1

II. RELATED WORK

Prior aerial manipulators face limits from coupled dynamics, constrained force directions, or controller switching. This work combines omnidirectional interaction with a simplified platform and describes its sensing and hardware configuration.

II. RELATED WORK

  • Underactuated manipulators have limited lateral forces and disturbance rejection, while prior fully actuated platforms still constrain some force directions or require additional manipulator degrees of freedom.The proposed approach combines omnidirectionality with high force and torque capabilities without an actuated manipulator arm.Source: §II · pp. 1–2
  • The proposed selective impedance strategy uses one model-based controller across situations, without transition handling, and plans motion relative to the local surface normal.Source: §II · p. 2
  • On-board visual-inertial sensing supports state estimation in complex environments, while a time-of-flight camera provides precise position and orientation relative to the environment.The paper identifies reliable state estimation near the environment as a challenge in moving systems from laboratory to field settings.Source: §II · p. 2

III. SYSTEM DESCRIPTION

  • The platform is a hexarotor with independently tilting double propeller groups, on-board computing and flight control, and batteries mounted to its base.Its main system parameters and coordinate frames are listed in Tables I and II.Source: §III · p. 2, Table I · p. 2, Table II · p. 2
  • A rigid arm carries the tool frame, with a nearby time-of-flight camera aligned to the tool and a visual-inertial sensor oriented oppositely for feature detection.Source: §III · p. 2, Figure 2 · p. 2

IV. CONTROL FRAMEWORK

The control framework transforms a reference wrench into rotor-speed and tilt-angle commands. Its simplified system model describes the six-dimensional motion and balances actuation with external forces and torques under stated modeling assumptions.

IV. CONTROL FRAMEWORK

  • The control framework allocates a reference wrench to rotor speeds u and tilt angles α.Actuator allocation follows.Source: §IV · p. 2

A. System Model

  • The model assumes a rigid body with principal-axis-aligned body axes, instantaneous thrust and drag response, negligible tilt-motor dynamics, and no propeller-group airflow interference.Thrust and drag torques are assumed proportional to squared rotor speeds.Source: A. System Model · pp. 2–3
  • The simplified system dynamics are derived in Lagrangian form, with v and its derivative stacking linear and angular velocity and acceleration.The model includes mass, centrifugal and Coriolis terms, gravity, and rotor-actuation and external force/torque vectors.Source: A. System Model · p. 3

B. External Wrench Estimation

The platform estimates external wrenches with a generalized-momentum approach. Differentiating the estimator relation yields a first-order low-pass estimate, requiring velocity estimates rather than acceleration measurements.

B. External Wrench Estimation

C. Interaction Control: Selective Impedance

The controller uses impedance dynamics to regulate interaction without contact detection or controller switching. Selective apparent inertia supports disturbance rejection in some directions and compliant behavior in others, with the rigid end-effector geometry determining the contact response.

C. Interaction Control: Selective Impedance

  • Impedance control regulates the exerted wrench without contact detection or controller switching, and also applies in free flight.Selective apparent inertia is used to reject disturbances in some directions while allowing compliance in others.Source: C. Interaction Control: Selective… · p. 3
  • The desired closed-loop dynamics use positive-definite matrices for apparent inertia, damping, and stiffness, acting on position and velocity errors.Source: C. Interaction Control: Selective… · p. 3
  • The impedance matrices are normalized with respect to system mass, and the selective impedance is rotated into the fixed end-effector frame.Source: C. Interaction Control: Selective… · p. 3
  • Lower apparent mass along the end-effector zt-axis makes aligning it with the surface normal sufficient for compliance in the contact direction and stiffness orthogonally.The same arrangement provides a stiff response to orientation error.Source: C. Interaction Control: Selective… · p. 3

D. Surface Normal and Distance Estimation

The system estimates a local contact point and surface normal from ToF-camera point clouds. It plans a target pose relative to that surface, then updates the surface estimate and trajectory while moving the set point tangentially.

D. Surface Normal and Distance Estimation

  • The contact point is the intersection of the observed surface and the tool zt-axis; nearby point-cloud samples provide its average location and a least-squares plane fit for the normal.The normal direction is flipped when it points away from the camera, ensuring consistency.Source: D. Surface Normal and Distance… · p. 3
  • In Fig. 4, C0 marks the contact point where the tool zt-axis intersects the observed surface.The figure is captioned as showing normal estimation and path planning.Source: D. Surface Normal and Distance… · p. 4, Figure 4 · p. 4

E. Path Planning in 6 DoF on a Manifold

  • The target pose is placed behind the estimated contact point and oriented with the tool zt-axis aligned to the estimated normal, supporting contact despite small state or distance-estimation fluctuations.The body yb-axis is aligned with the ground plane to constrain rotation about zt.Source: E. Path Planning in 6 DoF on a Manifold · p. 4
  • Translation along the surface moves the set point tangent to the estimated surface; the set point, surface estimate, and trajectory update at 5 Hz.This is described as pure-pursuit path tracking that maintains contact and surface orientation.Source: E. Path Planning in 6 DoF on a Manifold · p. 4, D. Surface Normal and Distance… · p. 4

A. Experimental Setup

The experiments use a tilt-rotor platform with external or onboard state estimation to evaluate disturbance rejection, surface interaction, onboard sensing, and concrete NDT contact testing.

A. Experimental Setup

  • Experiments use the tilt-rotor platform with the tool and impedance parameters listed in Table III; inertial parameters are multipliers of system inertia.Source: A. Experimental Setup · p. 4, Table III · p. 4
  • External state estimation fuses Vicon motion capture at 100 Hz with IMU measurements at 250 Hz.Source: A. Experimental Setup · p. 4
  • Onboard estimation combines Rovio visual-inertial position and attitude with ToF-based distance and surface-orientation estimates; only onboard data is processed in Section V-D.Source: A. Experimental Setup · p. 4
  • The experiments target wrench disturbance rejection, repeatable planar interaction, interaction with complex surfaces using onboard sensing, and infrastructure NDT testing.Source: A. Experimental Setup · p. 4

B. Rope Pull Disturbance in Free Flight

Rope-pull tests show that selective apparent inertia trades compliance against disturbance rejection, while whiteboard trials demonstrate stable, repeatable push-and-slide interaction despite friction and model error.

B. Rope Pull Disturbance in Free Flight

C. Push-and-Slide Along a Flat Surface

D. Depth Servoing for Contact with Unknown Surfaces

Using onboard sensing without prior maps or structural information, the system aligns its tool with local surface normals and traces a surface path on a ceiling vault.

D. Depth Servoing for Contact with Unknown Surfaces

  • The system autonomously approached an unknown ceiling vault, aligned the tool axis with the locally estimated normal, made contact, traced a line, and departed using onboard sensing alone.The contact site was approximately 4 m above ground in a staircase landing in daily use.Source: D. Depth Servoing for Contact with… · p. 6, Figure 8 · p. 6
  • During contact, the tool traversed 0.53 m along the surface at an average velocity of 0.17 m s−1.Source: D. Depth Servoing for Contact with… · p. 7
  • Surface-normal and distance estimates were smooth and consistent, supporting a clean target trajectory and reliable contact despite several centimeters of initial onboard-state-estimator drift.Frequent planning updates and drift-free normal and distance estimates made the system robust to that drift.Source: D. Depth Servoing for Contact with… · p. 7
  • During surface translation, the attitude change affected rotation only about the yt-axis while the tool attitude followed the estimated normal.Source: Figure 9 · p. 6, D. Depth Servoing for Contact with… · p. 7

E. Tests on Concrete Block with NDT Contact Sensor

Concrete-block tests show that the platform can maintain measurement contact and identify corrosion locations; estimator trials assess onboard pose and force estimates, with torque estimation left unevaluated.

E. Tests on Concrete Block with NDT Contact Sensor

  • During concrete contact, the system generated a 1.8 N surface-directed force while remaining robust to mass and center-of-mass model error in high-apparent-inertia directions.The specimen blocked xW motion during contact, while low apparent inertia along zt enabled compliance.Source: E. Tests on Concrete Block with NDT… · p. 7, Figure 10 · p. 7
  • NDT measurements indicated corrosion at contact points 2 and 3 on the concrete specimen.The sensor measured potential and resistance, which can indicate steel corrosion state; the specimen had a known corrosion spot.Source: E. Tests on Concrete Block with NDT… · p. 7, Figure 11 · p. 8

F. Evaluation of Force and State Estimators

  • Onboard state estimation adequately replaced motion capture for interactive flights of short duration, despite several-centimeter drift at initial contact and greater error during motion than quasi-static hovering.The system remained stable under impedance control despite a 0.6 m set-point offset.Source: F. Evaluation of Force and State… · p. 7, Figure 12 · p. 8
  • Estimated force magnitudes generally matched ground truth, but contact-direction error was 1.62 N, largely due to slow response to a step force.Further tuning of KI could improve estimator fidelity without causing instability.Source: F. Evaluation of Force and State… · pp. 7–8
  • Torque-estimation error was not evaluated because the sensor lacked a fixed connection, leaving it for future work.Source: F. Evaluation of Force and State… · p. 8

VI. CONCLUSION

The platform tracks full-pose trajectories during interaction while rejecting disturbances, and NDT-sensor experiments validate it for contact inspection. The authors identify estimator and localization issues to address before autonomous structural evaluation.

  • Experiments demonstrate full-pose trajectory tracking during interaction while rejecting disturbances with an impedance controller using selective apparent inertia.Source: §VI · p. 8
  • NDT-sensor experiments validate the platform as a tool for contact-based inspection.Source: §VI · p. 8
  • The wrench estimator does not distinguish wall and ground effects from contact forces, which may complicate interaction-force control; wind estimation could improve performance.Source: §VI · p. 8
  • Position and yaw drift in the VI state estimator motivates adding absolute localization; future work will integrate mapping and generate coverage trajectories for complete structural evaluation.Source: §VI · p. 8

Related papers

Loading 1905.03502v2…