⚡ electrical engineering

Gravity Compensation of the dVRK-Si Patient Side Manipulator based on Dynamic Model Identification

This paper addresses the control accuracy and response time challenges of the dVRK-Si Patient Side Manipulator caused by its upgraded structural gravity by presenting a novel full kinematic model and a dynamic model-based gravity compensation approach.

Original authors: Haoying Zhou, Hao Yang, Anton Deguet, Loris Fichera, Jie Ying Wu, Peter Kazanzides

Published 2026-01-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Haoying Zhou, Hao Yang, Anton Deguet, Loris Fichera, Jie Ying Wu, Peter Kazanzides

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a high-tech robotic arm used in surgery, specifically the "Patient Side Manipulator" (PSM) from the da Vinci Si system. Think of this arm as a very long, heavy crane. In older versions of this robot, the crane was so light that gravity didn't really bother it; it could just sit there without falling over. But the new version (the dVRK-Si) has been upgraded with heavier, sturdier parts. Now, gravity is a big deal. If you let go of the controls, the heavy arm wants to droop or swing down, just like a heavy door that doesn't have a spring to hold it open.

This paper is about teaching this heavy robotic arm how to "hold its own weight" automatically.

The Problem: The Heavy Arm

The researchers found that because this new robot arm is heavier, the force of gravity pulls on it constantly. If the robot's computer doesn't actively fight against this pull, the arm will drift away from where the surgeon wants it to be. It's like trying to balance a broom on your hand; if you aren't constantly making tiny adjustments, it falls. In a surgery, even a tiny drift is bad news.

The Solution: Learning the Arm's "Weight Map"

To fix this, the team didn't just guess how heavy the arm was. Instead, they treated the robot like a puzzle they needed to solve mathematically.

Building the Blueprint: First, they created a precise digital map (a kinematic model) of the robot. They measured every single link, joint, and motor, noting exactly how long they are and how they connect. It's like measuring every bone and muscle in a body to understand how it moves.
The "Shake and Tell" Test: To figure out exactly how the robot reacts to gravity and friction, they made the robot move along specific, calculated paths (excitation trajectories). Imagine shaking a box of LEGOs to figure out how heavy each piece is inside without opening the box. By watching how the robot moved and how much power the motors used, they could calculate the exact "dynamic model."
The Gravity Compensation: Once they knew exactly how heavy every part was and how the joints resisted movement (friction), they programmed the robot to apply the exact opposite force to counteract gravity. It's like a person holding a heavy backpack; the robot's computer calculates exactly how much muscle force is needed to keep the backpack from pulling them down, so the person doesn't have to think about it.

The Results: Holding Still

The team tested this by telling the robot to move to a specific spot and then switching off the "holding" motors, letting only the gravity-compensation system take over.

The Drift Test: They watched to see if the robot would "drift" (move on its own) over 5 seconds.
The Outcome: The robot stayed incredibly still. The new system was much better at keeping the arm in place than the old method. The robot didn't need to use maximum power to stay put; it used just the right amount of force, sitting comfortably in the "middle" of its capabilities.
Friction: They noted that one specific joint (the one that slides in and out) had a lot of "stickiness" (friction), which made it slightly harder to balance perfectly, but the system still worked very well.

In a Nutshell

The researchers successfully taught a heavy, upgraded surgical robot how to stand up straight on its own. By carefully measuring its parts and calculating exactly how gravity pulls on them, they created a system that automatically cancels out the weight. This means the robot can hold its position with high precision, making it a more reliable tool for future research and potential medical use, without needing constant, heavy-handed corrections from a human operator.

Technical Summary: Gravity Compensation of the dVRK-Si Patient Side Manipulator based on Dynamic Model Identification

Problem Statement
The da Vinci Research Kit (dVRK) is a critical open-source platform for advancing research in robot-assisted surgery. While the original dVRK Classic system utilized a Patient Side Manipulator (PSM) where gravity effects were negligible, the newer dVRK-Si system, derived from the da Vinci Si Surgical System, features a structurally upgraded PSM. This structural change results in a significant gravity offset that can no longer be ignored. Uncompensated gravity in the dVRK-Si PSM leads to reduced control accuracy and increased response times. Although dynamic model identification has been addressed for the dVRK Classic, the mechanical differences in the dVRK-Si necessitate new research into model-based control to achieve enhanced performance.

Methodology
The authors propose a comprehensive approach to gravity compensation (GC) for the dVRK-Si PSM, centered on identifying the full dynamic model parameters and isolating the gravity term, $G(q)$ , for compensation. The methodology proceeds through five distinct steps:

Kinematic Modeling: A full kinematic model of the dVRK-Si PSM was constructed based on manufacturer specifications and physical measurements. This model utilizes modified Denavit-Hartenberg (DH) parameters (detailed in Table I) and defines a specific frame structure (Fig. 2) to account for the complex linkage of the Si PSM, including the Remote Center of Motion (RCM).
Dynamic Modeling: The dynamic model was derived using the Euler-Lagrangian approach. The equations of motion are expressed as $M(q)\ddot{q} + C(q, \dot{q})\dot{q} + G(q) = \tau$ , where the goal is to identify the parameter vector $\delta$ representing link inertia, motor inertia, friction, and stiffness.
Excitation Trajectory Design: An optimal excitation trajectory was calculated based on joint constraints and the dynamic model to ensure rich data collection for parameter identification.
Data Collection: The excitation trajectory was executed on the physical dVRK-Si PSM to collect kinematic and dynamic data.
Parameter Identification: The dynamic parameters were solved using a convex optimization approach. To prevent overfitting and ensure physical validity, the optimization incorporated specific constraints: link masses must be positive, centers of mass must remain within the link's convex hull, and friction coefficients, motor inertia, and spring stiffness must be positive.

Key Contributions
The paper presents two primary contributions:

A novel full kinematic model of the dVRK-Si PSM, explicitly defining the frame relationships and parameters required for accurate modeling.
A gravity compensation strategy based on dynamic model identification. By identifying the full dynamic model, the authors extract the $G(q)$ term to provide the necessary torque for gravity compensation, rather than relying on heuristic or simplified models.

Results
The efficacy of the proposed gravity compensation was validated through a drift test. In this experiment, the PSM was moved to a desired pose using a standard PD position controller, then switched to an open-loop torque-based gravity compensation mode for 5 seconds. Drift was defined as movement exceeding 1 degree or 1 mm.

Drift Performance: The system maintained the robot's pose without noticeable drift across five random test configurations.
Error Reduction: The gravity compensation approach demonstrated smaller steady-state errors compared to the PD controller alone.
Torque Analysis: The joint torques generated by the GC method for the first three joints (which are most affected by gravity) fell within the calculated lower and upper bounds required to hold the joints in position against static friction. In most cases, these torques were close to the midpoint of the feasible range.
Joint 3 Observations: For the third joint (a prismatic joint), some deviation from the torque midpoint was observed, which the authors attribute to significant friction in the prismatic joint and the cannula.

Significance and Claims
The authors claim that the proposed gravity compensation approach effectively maintains the dVRK-Si PSM's pose without noticeable drift, addressing a critical limitation of the system's structural upgrade. The study demonstrates that model-based identification can successfully generalize across different dVRK-Si PSMs provided they share the same mounting-angle configuration. By enabling accurate gravity compensation, this work facilitates higher control accuracy and faster response times, thereby supporting more advanced research and control applications for the dVRK-Si platform. The work does not claim to solve all dynamic control challenges but specifically targets the gravity offset issue inherent to the Si PSM design.

The Problem: The Heavy Arm

The Solution: Learning the Arm's "Weight Map"

The Results: Holding Still

In a Nutshell

More like this