Preserving Operator Intent Under Actuation Constraints: Teleoperation of Manipulators with Coupled Transmissions

I. INTRODUCTION

• Independent motor saturation can alter the differential wrist’s pitch–roll ratio and motion direction, while infeasible commands may accumulate and trigger abrupt recovery motions.The proposed method jointly accounts for gravity compensation, joint mapping, and actuator constraints.Source: §I · pp. 1–2
• Actuator-space projection seeks to preserve local motion direction and pitch–roll coordination, while continuous feasible-reference updates limit command accumulation during saturation or short communication dropouts.The method extends an existing leader–follower hardware and real-time control pipeline.Source: §I · pp. 1–2

A. Immersive and Digital-Twin Teleoperation

• Immersive and digital-twin interfaces improve viewpoint control and operator intuitiveness, but often pass commands through direct joint tracking or Cartesian control without addressing actuator-level distortion.The proposed feasibility layer complements those interfaces when the follower reaches execution limits.Source: A. Immersive and Digital-Twin… · p. 2

B. Learning-Based Teleoperation and Shared Autonomy

• Learning-based teleoperation can support task-level behavior, but learned policies do not inherently guarantee safe execution under unseen actuator constraints or preserve instantaneous multi-DoF coordination.The proposed model-based approach operates before policy execution and requires no task-specific training data.Source: B. Learning-Based Teleoperation and… · p. 2

C. Optimization-Based and Constraint-Aware Robot Control

• Joint- or Cartesian-space constraint methods often treat actuator saturation as final clipping, which can change coupled-joint coordination and allow infeasible references to accumulate.The paper instead formulates coupled wrist motion and gravity-compensated actuator margins in actuator space.Source: C. Optimization-Based and… · p. 2

A. Problem Formulation and System Model

• The differential wrist maps physical pitch–roll coordinates to motor coordinates, so the two wrist degrees of freedom share actuator capacity and motor limits create coupled pitch–roll constraints.The manipulator uses a serial chain with a dual-motor differential wrist.Source: A. Problem Formulation and System Model · p. 2, Figure 2 · p. 3

B. Gravity Compensation and Actuator Feasibility

• The available actuator capability varies with robot configuration and payload after gravity compensation.The remaining motion torque is determined relative to the motor torque limits.Source: B. Gravity Compensation and Actuator… · p. 3
• Motor velocity constraints define which joint velocities are feasible for the differential wrist.The wrist’s motor-level limits apply jointly to pitch and roll.Source: B. Gravity Compensation and Actuator… · p. 3
• With equal transmission ratios and symmetric limits, the differential wrist’s feasible velocities satisfy |κp ˙qp| + |κr ˙qr| ≤ ¯v; for κp = κr, this forms a coupled diamond-shaped region.The region is not two independent joint intervals.Source: B. Gravity Compensation and Actuator… · p. 3

C. Actuator-Space Intent-Preserving Projection

• The projection obtains a feasible follower velocity because direct inverse-kinematics commands do not guarantee joint, transmission, or actuator feasibility.An execution-progress variable α indicates how much of the operator command is executed, with α = 1 denoting full execution.Source: C. Actuator-Space Intent-Preserving… · p. 3
• Figure 3 illustrates successful execution procedures for Insertion, Place, Follow, and Steer.Source: Figure 3 · p. 4
• When full execution is infeasible, the controller reduces execution speed before modifying the operator’s motion direction.The objective penalizes motion orthogonal to the command and preserves its parallel component.Source: C. Actuator-Space Intent-Preserving… · pp. 3–4
• The differential-wrist projection prevents independent motor clipping from changing the requested relative pitch–roll motion.Source: C. Actuator-Space Intent-Preserving… · p. 4

D. Continuous Feasible Reference

• Updating the follower reference with optimized feasible velocity avoids accumulating infeasible leader commands as position error that can cause catch-up motion.Source: D. Continuous Feasible Reference · p. 4
• For periodic encoder variables, the continuous state adds the shortest angular difference between successive measurements.The resulting state supports velocity estimation, continuous wrist-roll accumulation, and reference generation.Source: D. Continuous Feasible Reference · p. 4
• When feedback is unavailable or times out, the reference freezes and synchronizes with the measured configuration; after feedback returns, it restarts from that configuration.This avoids using an unexecuted historical command and suppresses delayed recovery transients.Source: D. Continuous Feasible Reference · p. 4

E. Low-Level Joint and Motor Control

• The control loop combines gravity feedforward with cascaded position–velocity tracking of the feasible reference.The stated torque command is τq = Kp(qref − q) + Kd(q̇ref − q̇) + g(q).Source: E. Low-Level Joint and Motor Control · p. 4
• For the differential wrist, physical pitch–roll torque is converted to corresponding motor torque and current using the motor torque constant.Source: E. Low-Level Joint and Motor Control · p. 4
• When gravity load or actuator saturation reduces capability, α decreases while preserving the operator’s local motion direction and differential-wrist coordination.With sufficient actuation, α remains close to one and the follower tracks the leader directly.Source: E. Low-Level Joint and Motor Control · pp. 4–5

A. Setup

• The real-system validation collected joint-position and velocity signals at both the controller and slave manipulator for comparison of operator input with execution.Source: A. Setup · p. 5
• The experiment used six joint channels and assessed continuity and coordination under differential-wrist coupling, actuator limits, and transient variations.Position records covered approximately 70.07 s on the controller and 123.75 s on the slave; velocity records covered approximately 58.00 s and 69.76 s, respectively.Source: A. Setup · p. 5
• Controller signals averaged about 9.82 Hz, while slave position and velocity signals averaged about 100.01 Hz.Source: A. Setup · p. 5

• Controller and manipulator velocity-channel means were both approximately 572 raw units, indicating comparable overall motion levels.Source: B. Result · p. 5
• Manipulator velocity standard deviations were 0.90, 1.66, and 1.26 raw units for Yaw, Pitch 1, and Pitch 2, while Pitch 3 and Roll 2 ranges reached 329 and 335 raw units.Source: B. Result · p. 5
• Motion remained continuous for most of the duration without evident sustained error accumulation.Source: B. Result · p. 5

1) Settings:

• Only the gravity-compensation configuration varied; master-slave mapping, differential-wrist control, actuator constraints, and continuous feasible-reference updates stayed fixed.Source: 1) Settings · p. 5

2) Result:

• 22.41 without gravity compensation fell to 5.91 with the full method in average velocity standard deviation, a 73.6% reduction.Source: 2) Result · p. 5, Figure 5 · p. 6

V. CONCLUSION

• The conclusion identifies joint design of gravity compensation, actuator-space constraints, and continuous feasible-reference updates as necessary.Source: §V · p. 6