Curtain Wall Installation Robotic Arm

• Curtain wall installation robotic arms are multi-DOF manipulators designed to precisely position heavy façade panels using hydraulic actuation, specialized end-effectors, and integrated BIM workflows.
• They employ advanced models that couple kinematics, hydraulic dynamics, and friction identification to optimize low-speed smoothness and ensure operational safety under challenging environmental conditions.
• Trajectory planning techniques, including biomimetic and multi-objective optimization methods, enable coordinated control of the arm and mobile base, improving energy efficiency and installation accuracy.

Searching arXiv for the specified curtain wall robotic arm papers and closely related work to ground the article. {"query":"cat:cs.RO OR cat:eess.SY curtain wall installation robotic arm hydraulic parameter identification trajectory planning biomimetic whole-body control", "max_results": 10} I found several directly relevant arXiv papers on curtain wall installation robotics, including dynamic parameter identification, biomimetic trajectory planning, multi-objective trajectory planning, and coordinated leg-arm control. I’ll synthesize these with precise claims from the provided data only. A curtain wall installation robotic arm is a multi-DOF robotic arm used to pick up large, heavy façade panels and place them precisely on building exteriors (Liu et al., 22 Jul 2025). In current research, the term encompasses not only the manipulator itself but also the actuation system, end-effector, sensing stack, trajectory generation, and, in some systems, the mobile or legged base that contributes to final panel pose adjustment. The application is defined by heavy and brittle workpieces, operation at height, constrained approach geometry near the façade, and stringent demands on safety, smoothness, and repeatable alignment; recent systems therefore couple arm mechanics with dedicated construction-task modeling rather than treating installation as generic factory pick-and-place (Liu et al., 16 Sep 2025).

1. Operational problem and task envelope

Curtain walls are large, heavy façade panels, often glass and aluminum, that must be positioned with high accuracy on high-rise buildings (Liu et al., 23 Jul 2025). Traditional installation relies on cranes, manual guiding, and workers operating close to edges and at height, which leads to low efficiency, safety risks, and variable quality (Liu et al., 23 Jul 2025). The robotic-arm approach is motivated by the need to improve safety, precision, and repeatability while reducing dependence on sustained manual handling (Liu et al., 22 Jul 2025).

The task envelope is broader than a simple lifting motion. A curtain wall installation robot must handle significant payloads, reach over building edges and along façades, position panels with high accuracy while avoiding collisions with the structure, and often operate under power limitations that make trajectory energy consumption relevant (Liu et al., 22 Jul 2025). A design-oriented analysis of industrial-arm scaling for curtain wall work states that the application requires payloads typically $50\text{–}150\ \text{kg}$, long reach, high stiffness under load, positioning tolerances in millimeters, and robust operation in outdoor, dusty, windy environments (Anumula et al., 10 Nov 2025). This makes façade robotics simultaneously a manipulation problem, a structural problem, and a site-integration problem.

A recurring simplification is to regard the curtain wall arm as a straightforward enlargement of an indoor serial manipulator. The available studies do not support that simplification. They instead emphasize high load-to-weight ratio, explicit treatment of heavy-panel dynamics, façade-specific end-effectors, and task-specific control formulations that prioritize low-speed smoothness, contact safety, and installation geometry (Liu et al., 23 Jul 2025).

2. Mechanical architectures and actuation regimes

Recent curtain wall systems exhibit several distinct arm architectures, all specialized for construction deployment rather than factory-cell operation.

System	Arm structure	Stated properties
Hydraulic foldable serial arm (Liu et al., 23 Jul 2025)	Foldable serial manipulator on a foot-track mobile base	Overall length 2.3 m; maximum payload 40 kg glass panels up to 1 m × 1 m
Hybrid folding + serial-parallel arm (Liu et al., 23 Jul 2025)	Folding arm plus serial-parallel arm with suction-cup end-effector	Six DOF; compliant operations over a 150° range; wrist motor with ±360° rotation
Hexapod-integrated installation arm (Liu et al., 16 Sep 2025)	3-DOF folding arm plus 2.3 m serial-parallel manipulator	Chassis payload 300 kg; folding-arm end-effector load 80–100 kg; installation height to 3 m

The hydraulic foldable serial arm reported in "Dynamic Parameter Identification of a Curtain Wall Installation Robotic Arm" is mounted on a foot-track mobile base, has an overall length of 2.3 m, supports 40 kg glass panels up to $1\ \text{m} \times 1\ \text{m}$, and is remotely operated with multi-angle visual monitoring (Liu et al., 23 Jul 2025). Its kinematic layout is a large folding arm in joints 1–3 plus a wrist in joints 4–6, with joint limits and speed/acceleration constraints explicitly tuned to façade reach and precise panel orientation (Liu et al., 23 Jul 2025).

The arm in "Multi-Objective Trajectory Planning for a Robotic Arm in Curtain Wall Installation" combines a folding arm for coarse positioning with a serial-parallel section near the tool for stiffness and load capacity (Liu et al., 23 Jul 2025). The paper describes a total of six DOF, hydraulically actuated primary joints, a suction-cup end-effector, compliant operations over a 150° range, and a wrist motor with ±360° rotation (Liu et al., 23 Jul 2025). This architecture is aimed at long reach with local stiffness near the panel interface.

A more integrated construction-robot variant is presented in "Leg-Arm Coordinated Operation for Curtain Wall Installation", where the manipulator is part of a hexapod platform (Liu et al., 16 Sep 2025). The chassis has six hydraulically actuated legs with 3 DOF each, supports up to 300 kg payload, and carries a 3-DOF folding arm with overall dimensions $$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$, plus a 2.3 m serial-parallel manipulator with a vertical range of 150° and suction-cup end-effector for panel pickup from the ground and transport to wall, ceiling, or floor placement (Liu et al., 16 Sep 2025). In this configuration, body pose adjustment by the legs is treated as part of the installation arm’s effective workspace.

Hydraulic actuation is dominant in the curtain wall-specific systems. In the 2.3 m foldable serial arm, all joints are driven by hydraulic cylinders, with pressure and displacement sensing, multi-channel proportional valve modules, and explicit modeling of cylinder force balance, friction, and pressure effects (Liu et al., 23 Jul 2025). The rationale is high power density and load handling. A plausible implication is that curtain wall arm design is less about choosing between “robotic” and “construction” mechanisms than about embedding manipulator kinematics inside a hydraulically constrained machine architecture.

Related transmission research points to an alternative design direction. "D3-ARM" presents a fully decoupled cable-driven six-DOF arm with a 776 mm moving part, 1.6 kg moving mass, base-mounted actuators, and 2.0 kg payload capacity (Luo et al., 18 Feb 2025). The reported payload is far below curtain wall panel handling, but the study identifies low moving mass, protected base-mounted motors, and decoupled tendon routing as properties that are attractive in harsh environments (Luo et al., 18 Feb 2025). For curtain wall work, this is a transmission concept rather than a complete panel-handling solution.

3. Kinematic, dynamic, and friction models

Curtain wall installation arms are modeled with standard rigid-body robotics formalisms, but recent work couples those formalisms to hydraulic actuator dynamics and low-speed friction models. In the 2.3 m hydraulic arm, the kinematic chain is defined by standard Denavit–Hartenberg parameters, with forward kinematics

$${}^{0}T_6(q) = {}^{0}T_1 {}^{1}T_2 \cdots {}^{5}T_6,$$

and rigid-body dynamics are written using iterative Newton–Euler as

$$\boldsymbol{\tau} = \mathbf{M}(\mathbf{q}) \ddot{\mathbf{q}} + \mathbf{C}(\mathbf{q}, \dot{\mathbf{q}})\dot{\mathbf{q}} + \mathbf{G}(\mathbf{q}) + \boldsymbol{\tau}_{f}(\dot{\mathbf{q}}).$$

The full model is recast into a linear-in-parameters regressor form, initially with 78 parameters, including inertial terms and friction terms (Liu et al., 23 Jul 2025).

The distinctive feature of this work is explicit integration of hydraulic cylinder dynamics into the robot model. Cylinder force is written as

$F_{\text{cyl}} = p_1 A_1 - p_2 A_2,$

and joint torque as

$\tau_i = J_{F\to\tau, i}(q)\, F_{\text{cyl},i}.$

Cylinder dynamics are described by

$$m\ddot{x}+c\dot{x}+Kx+F_d = p_1 A_1 - p_2 A_2 - F_L,$$

so that measured pressures and piston motion are mapped into joint torque estimates (Liu et al., 23 Jul 2025). This is central for heavy-panel placement because low-speed hydraulic behavior directly affects stick–slip, overshoot, and force predictability near the façade.

Friction is modeled with a Stribeck formulation. The paper gives a general Stribeck form and then linearizes it to

$$F_d(\dot{x}, f_c, f_v, f_s) = f_c \operatorname{sgn}(\dot{x}) + f_v \dot{x} + f_s \dot{x}^{1/3},$$

which preserves parameter linearity for least-squares identification (Liu et al., 23 Jul 2025). The same study reports that identified Stribeck curves differ across joints, including nominally similar joints, and gives examples such as $f_{c,1} = 20.77\ \text{N·m}$, $1\ \text{m} \times 1\ \text{m}$0, $1\ \text{m} \times 1\ \text{m}$1 for joint 1, and $1\ \text{m} \times 1\ \text{m}$2, $1\ \text{m} \times 1\ \text{m}$3, $1\ \text{m} \times 1\ \text{m}$4 for joint 6 (Liu et al., 23 Jul 2025).

Parameter identification is organized as a hierarchical progressive strategy. Cylinder parameters are first identified under unloaded sinusoidal piston excitation,

$1\ \text{m} \times 1\ \text{m}$5

followed by rigid-body parameter identification using truncated Fourier-series joint trajectories with $1\ \text{m} \times 1\ \text{m}$6 and $1\ \text{m} \times 1\ \text{m}$7 (Liu et al., 23 Jul 2025). SymPybotics is used to generate the regressor and reduce the 78-parameter model to 18 independent parameters (Liu et al., 23 Jul 2025). Experimental validation reports residual standard deviations below $1\ \text{m} \times 1\ \text{m}$8 for all six joints, with values of 0.226, 0.289, 0.378, 0.1929, 0.1787, and 0.3356 Nm for joints 1–6 respectively (Liu et al., 23 Jul 2025).

The broader significance is that curtain wall arm modeling is no longer limited to kinematics and nominal payload calculations. It increasingly uses composite models in which joint torque, hydraulic transmission, and friction identification are treated as prerequisites for feedforward torque computation, adaptive control, and safe low-speed façade interaction.

4. Trajectory planning, energy shaping, and whole-body control

Trajectory planning for curtain wall installation has developed along at least three lines: human-inspired energy shaping, explicit multi-objective optimization, and whole-body coordination between arm and base.

The biomimetic approach in "Trajectory Planning of a Curtain Wall Installation Robot Based on Biomimetic Mechanisms" starts from shoulder–elbow lifting biomechanics during dumbbell curls (Liu et al., 22 Jul 2025). The paper models the human upper limb as a 2-DOF planar chain, uses MediaPipe kinematics and EMG timing to divide lifting into three phases, and maps the pattern to a curtain wall arm so that it accelerates in the high-load capacity region, carries momentum through the weakest load region, and decelerates near the goal (Liu et al., 22 Jul 2025). Joint trajectories are parameterized as quintic polynomials and optimized with PSO so that the velocity peak occurs at approximately $1\ \text{m} \times 1\ \text{m}$9 elbow angle (Liu et al., 22 Jul 2025). The paper reports an energy consumption reduction of about 12% in the anthropomorphic trajectory-planning stage and a 48.4% reduction in the peak of the energy consumption curve in the full curtain wall robot simulation (Liu et al., 22 Jul 2025).

A second line is explicit many-objective optimization. "Multi-Objective Trajectory Planning for a Robotic Arm in Curtain Wall Installation" parameterizes joint motion with sixth-order B-splines and minimizes total time,

$$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$0

joint impact based on jerk,

$$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$1

and energy consumption based on mechanical power,

$$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$2

subject to torque, jerk, and velocity constraints (Liu et al., 23 Jul 2025). The paper proposes NSGA-III-FO, an NSGA-III variant with a focused operator screening mechanism, and reports better IGD values than NSGA-III, MOEA/D, and MSOPS-II on DTLZ3 and WFG3, including 341.3 ± 25.3 on DTLZ3 and 0.6087 ± 0.028 on WFG3 (Liu et al., 23 Jul 2025). Experiments on a hydraulic platform using a 10 kg steel plate validate vertical-surface and overhead installation tasks; the authors note larger tracking errors in Task 1 during transitions from static to dynamic phases and smaller errors in Task 2 under lower joint velocities (Liu et al., 23 Jul 2025).

The third line is whole-body coordination. In the hexapod system, the arm is not controlled in isolation. The paper describes a hierarchical optimization-based whole-body control framework in which stability and contact constraints have highest priority, end-effector tracking is next, and body posture and torque minimization are lower-priority objectives (Liu et al., 16 Sep 2025). The robot can grasp a panel from the ground, walk while holding it, and adjust installation angle and pose while moving (Liu et al., 16 Sep 2025). This suggests that for uneven terrain or edge-adjacent work, panel pose adjustment by the base may be as important as wrist dexterity.

Taken together, these planning studies show that curtain wall robotics has moved beyond point-to-point interpolation. Time, jerk, energy, and support-base coordination are treated as co-equal planning variables because panel fragility, hydraulic nonlinearities, and site constraints penalize abrupt motion more severely than in conventional industrial manipulation.

5. Sensing, BIM integration, fixation tooling, and collaborative workflows

A curtain wall installation robotic arm is increasingly embedded in a larger information and sensing architecture. One branch of this development links 4D BIM to robot task planning. "Integration of 4D BIM and Robot Task Planning" defines robot tasks as ordered sequences of actions linked to reusable robot skills, derives robot-related BIM augmentation requirements from those skills, and converts the BIM model into a robot simulation world in ROS/Gazebo (Oyediran et al., 2024). Although the case study is indoor wall frame installation, the task/action/skill representation directly generalizes to façade installation sequences such as navigation to panel storage, pick, transport, align, and install (Oyediran et al., 2024).

A closely related modular-robot framework extends this into morphology synthesis and execution. "Holistic Construction Automation with Modular Robots" uses BIM-based mission specification, CoBRA task representation, multi-objective genetic optimization of robot morphology, and ArUco-based calibration to compensate base placement errors during drilling (Külz et al., 2024). The paper formulates goals as $$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$3, where $$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$4 and $$950\,\text{mm} \times 230\,\text{mm} \times 1010\,\text{mm}$$5 is a tolerance function, and evaluates robustness through re-planning under bounded base-offset perturbations (Külz et al., 2024). For curtain wall arms, the same structure can be used to encode panel placement poses, approach offsets, and façade obstacles.

On-site installation also interacts with bracket fixation and human-supervised preparation tasks. "Dual-Arm Construction Robot for Automatic Fixation of Structural Parts to Concrete Surfaces in Narrow Environments" presents a two-arm system in which one arm holds the part and the other drills, inserts anchors, and tightens nuts with custom tools designed to reduce reaction forces on small industrial arms (Yasutomi et al., 2024). The reported system uses two 13 kg-payload Denso arms, force/torque sensing, laser distance sensing, and camera-based hole detection, and completes one-side fixation of a structural part in 9 min 28 s (Yasutomi et al., 2024). For curtain wall systems, these fixation operations are directly relevant at the bracket and anchor interface even when panel positioning is performed by a separate higher-payload arm.

Human-robot collaboration research supplies a further layer. "Adaptive Human-Robot Collaboration for Masonry Construction Under Material and Assembly Uncertainty" combines an end-effector-mounted projector with laser scanning so that spatially registered visual cues guide a human task and feedback-driven pose correction compensates for as-built deviation (Gao et al., 18 May 2026). The paper reports that projection guidance improves adhesive application consistency and reduces application time, while laser-based correction maintains level courses and avoids collision-prone failures associated with open-loop execution (Gao et al., 18 May 2026). A plausible implication is that analogous projector- and scan-based workflows could guide sealant placement, shim positioning, or bracket preparation in curtain wall installation.

Finally, outdoor mobile manipulation research shows how a global–local perception split can be organized in construction-like terrain. "Autonomous, Mobile Manipulation in a Wall-building Scenario" uses Google Cartographer for global mapping, Move Base with TEB for navigation, visual servoing for local object approach, and a passively compliant end-effector with tactile feedback for final grasp stabilization (Vatavuk et al., 2022). Although demonstrated on brick handling, the combination of coarse localization, local visual servoing, and passive compliance maps naturally onto façade operations in which the arm must first reach a bay and then perform sensor-guided micro-alignment.

6. Limitations, open problems, and likely directions

Current curtain wall installation robotic arms remain constrained by modeling simplifications and limited deployment conditions. In the hydraulic parameter-identification study, valve dynamics, full pressure dynamics, hose flexibility, and structural flexibility of long links are not fully identified; cylinder identification is performed unloaded, and experiments are conducted in controlled conditions rather than realistic construction sites with wind, temperature changes, and dust (Liu et al., 23 Jul 2025). The same work stops at model identification and validation, leaving model predictive control, adaptive control, and force-controlled panel installation as future directions (Liu et al., 23 Jul 2025).

The biomimetic planning study is validated in simulation rather than on physical façade hardware, does not integrate vision or real building environments, and uses EMG for pattern extraction rather than real-time adaptive control (Liu et al., 22 Jul 2025). The whole-body hexapod study demonstrates simultaneous walking and installation qualitatively, but does not report detailed positioning accuracy, disturbance-rejection metrics, or exteroceptive perception such as vision or LiDAR-based target localization (Liu et al., 16 Sep 2025). Related modular BIM-to-execution frameworks demonstrate that robust compensation for base error is possible, but their reported drilling precision is on the order of a centimeter rather than the millimeter-level repeatability often associated with finished façade alignment (Külz et al., 2024).

Several open problems therefore recur across the literature. One is the need to integrate as-built perception into the arm-level control loop, not only into offline task specification. Another is the treatment of payload as a variable dynamic entity rather than an externally imposed mass: the hydraulic identification work explicitly identifies panel mass and inertia as payload parameters that could be integrated into future models (Liu et al., 23 Jul 2025). A third is the coupling between panel handling and fixation. The dual-arm fixation results indicate that reaction-force management through custom tooling can make drilling and tightening feasible with smaller arms (Yasutomi et al., 2024); this suggests that future curtain wall systems may separate heavy panel support from high-precision fixation instead of demanding both from a single manipulator.

A further misconception is that better trajectory planning alone will make curtain wall robotics practical. The combined evidence indicates that practical systems require co-design of arm architecture, hydraulic or transmission model, friction identification, perception, installation workflow, and site logistics. This suggests that the mature curtain wall installation robotic arm will be a composite system: a panel-positioning mechanism, a fixation or contact-management subsystem, and a BIM- and sensing-driven planning layer that continuously reconciles nominal design geometry with as-built construction conditions.