Fast and Reliable Autonomous Surgical Debridement with Cable-Driven Robots Using a Two-Phase Calibration Procedure

Daniel Seita, Sanjay Krishnan, Roy Fox, Stephen McKinley, John F. Canny, Ken Goldberg

Abstract

Automating precision subtasks such as debridement (removing dead or diseased tissue fragments) with Robotic Surgical Assistants (RSAs) such as the da Vinci Research Kit (dVRK) is challenging due to inherent non-linearities in cable-driven systems. We propose and evaluate a novel two-phase coarse-to-fine calibration method. In Phase I (coarse), we place a red calibration marker on the end effector and let it randomly move through a set of open-loop trajectories to obtain a large sample set of camera pixels and internal robot end-effector configurations. This coarse data is then used to train a Deep Neural Network (DNN) to learn the coarse transformation bias. In Phase II (fine), the bias from Phase I is applied to move the end-effector toward a small set of specific target points on a printed sheet. For each target, a human operator manually adjusts the end-effector position by direct contact and the residual compensation bias is recorded. This fine data is then used to train a Random Forest (RF) to learn the fine transformation bias. Subsequent experiments suggest that without calibration, position errors average 4.55mm. Phase I can reduce average error to 2.14mm and the combination of Phase I and Phase II reduces average error to 1.08mm. We apply these results to debridement of raisins and pumpkin seeds as fragment phantoms. Using an endoscopic stereo camera with standard edge detection, experiments with 120 trials achieved average success rates of 94.5%, exceeding prior results with much larger fragments (89.4%) and achieving a speedup of 2.1x, decreasing time per fragment from 15.8 seconds to 7.3 seconds.

Introduction

Surgical robots like the da Vinci Research Kit (dVRK) are characterized by their cable-driven mechanisms. While this design provides necessary compliance and safety for surgical environments, it introduces significant non-linearities, including cable stretch, hysteresis, and friction. These factors complicate the task of precise end-effector localization based solely on internal sensors. For autonomous tasks such as surgical debridement—the removal of diseased tissue—millimeter-level precision is critical to avoid damaging healthy tissue. This research presents a "fast and reliable" calibration framework that addresses these non-linearities without requiring expensive external motion-capture systems.

Methodology

The proposed system utilizes a two-phase coarse-to-fine calibration strategy to map the relationship between the camera's visual field and the robot's internal coordinate system.

Overview of the two-phase calibration pipeline and its application to surgical debridement.

Phase I: Coarse Calibration (DNN)

In the first phase, the robot's end-effector is marked with red tape and programmed to follow automated, random trajectories within its workspace. An endoscopic camera tracks this marker using HSV thresholding, generating a dataset of 1,939 valid points mapping camera pixels to robot joint orientations. A Deep Neural Network (DNN) is trained on this data to learn the global, coarse transformation between the camera view and the robot base frame.

Phase II: Fine Calibration (RF)

To account for local residuals that the DNN might miss, a human-in-the-loop approach is employed. The robot attempts to reach the centers of a 35-point grid on a printed calibration sheet. A human operator then manually adjusts the robot tip to the precise center of each target. These small corrections, or residuals, are used to train a Random Forest (RF) regressor. By stacking these models—using the RF to correct the output of the DNN—the system achieves significantly higher accuracy. To further refine the model, the robot's yaw is discretized into five specific angles, with specialized models trained for each to account for tension changes at different tool rotations.

Experimental Results

Accuracy Improvement

The effectiveness of the two-phase approach was measured by the reduction in average positioning error:

  • No Calibration: 4.55 mm
  • Phase I (DNN) Only: 2.14 mm
  • Phase I + II (DNN + RF): 1.08 mm

Debridement Performance

The calibrated system was tested on fragment phantoms, specifically pumpkin seeds (slippery) and raisins (sticky). Across 120 trials, the system achieved a 94.5% success rate. This performance represents a significant improvement over previous autonomous methods, which achieved 89.4% success on larger fragments. Furthermore, the system demonstrated a 2.1x speedup, reducing the average time per fragment from 15.8 seconds to 7.3 seconds.

Discussion and Future Work

This work demonstrates that a hybrid modeling approach combining deep learning for global mapping and random forests for local error correction can overcome the inherent limitations of cable-driven surgical robots. The use of low-cost color-based tracking makes this method highly accessible for existing surgical setups.

Future research will focus on automating Phase II to remove the need for manual human adjustment. Additionally, while this study utilized a static environment, real-world surgical applications will require the integration of this calibration method with real-time visual servoing to handle dynamic, deforming tissues.

Summary updated

Abstract

Automating precision subtasks such as debridement (removing dead or diseased tissue fragments) with Robotic Surgical Assistants (RSAs) such as the da Vinci Research Kit (dVRK) is challenging due to inherent non-linearities in cable-driven systems. We propose and evaluate a novel two-phase coarse-to-fine calibration method. In Phase I (coarse), we place a red calibration marker on the end effector and let it randomly move through a set of open-loop trajectories to obtain a large sample set of camera pixels and internal robot end-effector configurations. This coarse data is then used to train a Deep Neural Network (DNN) to learn the coarse transformation bias. In Phase II (fine), the bias from Phase I is applied to move the end-effector toward a small set of specific target points on a printed sheet. For each target, a human operator manually adjusts the end-effector position by direct contact (not through teleoperation) and the residual compensation bias is recorded. This fine data is then used to train a Random Forest (RF) to learn the fine transformation bias. Subsequent experiments suggest that without calibration, position errors average 4.55mm. Phase I can reduce average error to 2.14mm and the combination of Phase I and Phase II can reduces average error to 1.08mm. We apply these results to debridement of raisins and pumpkin seeds as fragment phantoms. Using an endoscopic stereo camera with standard edge detection, experiments with 120 trials achieved average success rates of 94.5%, exceeding prior results with much larger fragments (89.4%) and achieving a speedup of 2.1x, decreasing time per fragment from 15.8 seconds to 7.3 seconds. Source code, data, and videos are available at https://sites.google.com/view/calib-icra/.
BibTeX
@inproceedings{seita2018fast,
  author = {Daniel Seita and Sanjay Krishnan and Roy Fox and Stephen McKinley and John F. Canny and Ken Goldberg},
  title = {{Fast and Reliable Autonomous Surgical Debridement with Cable-Driven Robots Using a Two-Phase Calibration Procedure}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2018},
  url = {https://openreview.net/forum?id=yySegmSul7A},
}
Figure