Aerial Continuum Manipulator (AeCoM)

• AeCoM is defined as a quadrotor UAV integrated with a tendon-driven continuum arm that offers high compliance and dexterity for manipulation in constrained environments.
• It employs constant-curvature kinematic models and sensor fusion from IMUs and load cells to ensure precise force estimation and robust tendon tension control.
• Experimental validations demonstrate superior payload handling, an expanded reachable workspace, and enhanced safety compared to traditional rigid-link aerial manipulators.

An Aerial Continuum Manipulator (AeCoM) is an aerial robotic platform that integrates a tendon-driven continuum arm with a rotary-wing unmanned aerial vehicle (UAV), most typically a quadrotor. Unlike classic aerial manipulators based on discrete, rigid-link serial kinematics, the AeCoM concept leverages the high dexterity and compliance of continuum robot arms and unifies mechanical design, kinematic modeling, and force/tension control to enable manipulation in cluttered, constrained, or safety-critical environments. This design enables tasks involving delicate grasping, interaction with soft or variable payloads, and engagement in confined spaces, while actively addressing key issues such as tendon slacking and force/moment transfer to the airborne base (Peng et al., 2021, Zhang et al., 2022, Zhao et al., 2022).

1. System Architecture and Mechanical Structure

The AeCoM paradigm centers on a lightweight, cable/tendon-driven continuum arm that is physically and functionally integrated with an aerial vehicle.

General Architecture:

• The system comprises a quadrotor (or equivalent rotary-wing UAV platform) as the mobile base, typically carrying a compact actuation module and the continuum manipulator beneath the main fuselage. Key features include:
  • Four-propeller platform (e.g., T-motor F90, with PixRacer controller and onboard computer)
  • Continuum arm, actuated via four tendon motors placed near the arm's root for minimal base moment
  • Two-rod folding landing gear for deployment flexibility

Manipulator Design:

• The continuum arm is segmented. In (Peng et al., 2021), the arm consists of five serial bending segments (each $L_s = 20\,\mathrm{mm}$ in length, $D \approx 24\,\mathrm{mm}$ outer diameter with 40% mass reduction cutouts), joined via 3D-printed gimbals and preloaded with four stiff springs ($k \approx 5\,\mathrm{N/mm}$) at radius $r=6\,\mathrm{mm}$.
• Cable routing utilizes strength-optimized Dyneema tendons anchored at the distal end disk; pulleys embedded in each disk guide the tendons to enforce orthogonally paired tendon actuation.
• Total arm mass is typically less than $50\,\mathrm{g}$, but higher-mass (e.g., $150\,\mathrm{g}$ arms with $420\,\mathrm{g}$ total module including actuation unit) are also reported for modular systems (Zhang et al., 2022, Zhao et al., 2022).
• Tendon actuation is achieved via miniature servo or dedicated motors with integrated torque sensing and electronic braking for slack prevention.

Structural Characteristics:

• The compliant backbone is typically NiTi (nickel-titanium), with precise characterization of Young’s modulus ($E_p \sim 82\,\mathrm{GPa}$) and second moment of inertia ($I_p$) for accurate modeling.
• Modular configurations are feasible by stacking continuum segments, providing scalable workspace and DOF (Zhang et al., 2022).

2. Kinematic and Static Modeling

Kinematic Models:

• The AeCoM leverages the constant-curvature (CC) or piecewise-constant curvature (PCC) model for its workspace mapping and control.
• The backbone shape in a single segment is parameterized by curvature $\kappa = \theta/L$, where $D \approx 24\,\mathrm{mm}$0 is the total bending angle and $D \approx 24\,\mathrm{mm}$1 the segment length; the bending plane orientation $D \approx 24\,\mathrm{mm}$2 determines the spatial pose.
• Forward kinematics:

  $D \approx 24\,\mathrm{mm}$3

• The configuration–joint mapping relates tendon length changes $D \approx 24\,\mathrm{mm}$4 or actuator displacements $D \approx 24\,\mathrm{mm}$5 to $D \approx 24\,\mathrm{mm}$6 via

  $D \approx 24\,\mathrm{mm}$7

  with inverse solutions exploiting the orthogonal actuation geometry.

Task Space and Jacobian Formulation:

• The manipulator’s geometric Jacobian maps shape rates to end-effector twist, supporting both kinematic and compliance control frameworks. The explicit $D \approx 24\,\mathrm{mm}$8 and $D \approx 24\,\mathrm{mm}$9 forms relate workspace and joint actuation for both feedforward manipulation and force estimation (Zhang et al., 2022, Zhao et al., 2022).

Statics and Stiffness:

• Elastic energy for a continuum section is

  $k \approx 5\,\mathrm{N/mm}$0

• Equilibrium equations balancing tendon-generated forces ($k \approx 5\,\mathrm{N/mm}$1) and external wrench ($k \approx 5\,\mathrm{N/mm}$2) with internal elastic response:

  $k \approx 5\,\mathrm{N/mm}$3

• The configuration-space stiffness matrix $k \approx 5\,\mathrm{N/mm}$4 and its mapping to task-space stiffness ($k \approx 5\,\mathrm{N/mm}$5) facilitate compliance analysis and design trade-offs.

3. Integrated Sensing, Force Estimation, and Tension Control

Sensor Fusion for Kinematics:

• Tension in each tendon is measured via inline load cells; position or orientation at the end disk is measured by an IMU (e.g., MPU9250 or BNO080). The IMU provides $k \approx 5\,\mathrm{N/mm}$6 for orientation correction.
• Sensor fusion is performed via a complementary filter, combining IMU-based tip orientation with PCC-model predictions to mitigate modeling errors, especially under variable payload (Peng et al., 2021).

Tendon Slack Prevention:

• Closed-loop tension control is implemented for each actuator, using a PI controller:

  $k \approx 5\,\mathrm{N/mm}$7

  where $k \approx 5\,\mathrm{N/mm}$8 sets nominal pretension.

• Each actuator incorporates an electronic brake, engaging rapidly when measured tension drops below threshold, thus preventing slack during aggressive flight or manipulation motions.

IMU-based Force Estimation:

• The end-disk IMU enables onboard estimation of tip force:
  • Eg., inward tip force average estimation error: $r=6\,\mathrm{mm}$1 ($r=6\,\mathrm{mm}$2) (Zhang et al., 2022).

4. Experimental Validation and Performance

Payload Handling and Dexterity:

• The AeCoM demonstrates successful manipulation of $r=6\,\mathrm{mm}$3 cylindrical objects (8× arm mass), with EE position error of $r=6\,\mathrm{mm}$4 RMS over ten trials (Peng et al., 2021).
• Workspace: bending allows a continuous reachable volume $r=6\,\mathrm{mm}$5 (vs. $r=6\,\mathrm{mm}$6 for equivalent 2-DOF rigid arms).

Kinematic and Force Tracking:

• Dynamic trajectory following in bending ($r=6\,\mathrm{mm}$7) space at $r=6\,\mathrm{mm}$8 achieves EE path errors $r=6\,\mathrm{mm}$9 RMS.
• Under $50\,\mathrm{g}$0–$50\,\mathrm{g}$1 payload variation, IMU fusion reduces orientation error from $50\,\mathrm{g}$2 to $50\,\mathrm{g}$3 RMS, and position error from $50\,\mathrm{g}$4 to $50\,\mathrm{g}$5.

Tendon Tension and Slack Prevention:

• Under fast bending ($50\,\mathrm{g}$6), tendon tension varies by $50\,\mathrm{g}$7 with no slack events under feedback; slack occurred 18% of the time without active tension control.
• Measured tendon tension mean: $50\,\mathrm{g}$8 during full-range motion; standard deviation of slack-related events is zero (Peng et al., 2021).
• Compliance during interaction (mock "perching" scenarios) reduces peak contact forces by approximately 40% compared to rigid arms (Zhang et al., 2022, Zhao et al., 2022).

Stiffness and Compliance Tests:

• Radial tip loading deflections are configuration-dependent: at $50\,\mathrm{g}$9, tip deflects $150\,\mathrm{g}$0 at $150\,\mathrm{g}$1; at $150\,\mathrm{g}$2, deflection is $150\,\mathrm{g}$3. Increasing the bend increases the local stiffness (Zhao et al., 2022).

5. Comparative Analysis and Application Domains

Advantages Over Rigid-Link Manipulators:

• Payload ratio: arm mass to lifted mass is $150\,\mathrm{g}$4 (AeCoM) vs. $150\,\mathrm{g}$5 (rigid-link).
• Safety: compliant structure reduces collision impact forces by 60%.
• Workspace: significantly larger continuous bending workspace, enabling complex tasks in restricted geometries.

Deployment and Use-Case Scenarios:

• Confined-space inspection: ability to snake into ducts or interstitial environments.
• Aerial assembly: compliance aids in tight-insertion maneuvers.
• Delicate grasping: soft tendon routing and compliance reduce object damage risk.

Integration Considerations:

• Mass and inertia of the manipulator base are co-located for minimal UAV disturbance.
• Control architectures recommend cascaded outer-vehicle/inner-manipulator loops, explicit compensation of base reaction forces, and impedance-style tip interaction for force-critical tasks (Zhao et al., 2022).

6. Limitations and Research Directions

Identified Limitations:

• Active-stiffness contributions from tendon pre-tensioning need explicit inclusion for high-accuracy force estimation under outward loadings.
• Mechanical backlash in cable routing can degrade positioning accuracy.
• Some studies report only mock-flight validation; full-flight dynamic tests remain to be extensively documented (Zhang et al., 2022, Zhao et al., 2022).

Open Research Problems:

• Quantifying payload–energy cost tradeoffs in dynamic aerial operation.
• Extending to multi-segment continuum arms for 3D, S-shaped workspace and advanced collision avoidance.
• Robust online estimation of arm configuration (potentially integrating curvature sensors or vision-based shape reconstruction).
• Development of fully coupled UAV+continuum arm controllers with dynamic compensation (Coriolis, centrifugal, aerodynamic coupling).
• Enhanced tip sensory feedback for improved manipulation in unstructured or contact-rich tasks.

The AeCoM framework establishes a foundational architecture for compliant, dexterous, and force-aware aerial manipulation based on lightweight tendon-driven continuum rods, sensor-integrated kinematics, and robust tension/force control, with validated improvements in workspace, safety, and handling capacity relative to conventional rigid-link aerial manipulators (Peng et al., 2021, Zhang et al., 2022, Zhao et al., 2022).