⚡ electrical engineering

Research on Adaptive Fuzzy PID Control of Breast Intervention Robot Based on Visual Detection

This paper proposes an adaptive fuzzy PID control system integrated with a real-time binocular stereo vision workflow to achieve sub-millimeter needle tip tracking and significantly improve the targeting accuracy of breast intervention robots for clinical applications.

Original authors: haiyan du, zhimeng zhang, xilei wang, guijin xie

Published 2026-08-11

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: haiyan du, zhimeng zhang, xilei wang, guijin xie

Original paper licensed under CC BY 4.0 (https://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 world where robots don't just build cars or vacuum floors, but help doctors perform delicate surgeries. This is the exciting corner of science known as medical robotics. At its heart, these machines need two superpowers: vision and control. Think of vision as the robot's eyes, allowing it to see exactly where a needle is in 3D space, much like how your brain combines what both your eyes see to judge distance. Then there's control, the robot's "brain" for movement. A standard control system is like a driver who only looks at the speedometer; if they go too fast, they slam on the brakes, often overshooting the stop sign. A smarter system, however, is like a skilled driver who feels the road, anticipates bumps, and gently adjusts the steering and gas to stop exactly where they need to. Why does anyone care? Because in breast cancer treatment, doctors need to guide a needle into a tiny tumor with extreme precision. Even a small wobble or a missed target can mean the difference between a successful biopsy and unnecessary harm.

The Robot That Learns to Steady Its Hand

In this study, a team of researchers from Harbin University of Science and Technology tackled the tricky problem of making a breast intervention robot move with the steadiness of a master surgeon. They wanted to solve a specific headache: when a robot tries to push a needle into soft, squishy breast tissue, it often shakes or overshoots its target. To fix this, they built a system that combines binocular stereo vision (two cameras working together like human eyes) with a clever new way of controlling the robot's motors called adaptive fuzzy PID.

The Eyes: Seeing the Needle Tip in 3D

First, the team had to teach the robot how to see. They didn't just use one camera; they used a pair, creating a "binocular" setup. Imagine looking at a finger with one eye closed, then the other; your brain notices the finger shifts position. The robot does the same thing. By using two cameras, it can calculate exactly where the tip of the biopsy needle is in 3D space.

The process is like a high-tech magic trick involving several steps:

Cleaning the Picture: The raw images from the cameras are a bit messy, so the computer first smooths them out, removing "noise" (like static on an old TV).
Aligning the View: The two camera images are mathematically "straightened" so they line up perfectly, like aligning two pieces of a puzzle.
Finding the Match: The computer looks for the same point on the needle tip in both images. It uses a method called SGBM (Semi-Global Block Matching) to find the best match, creating a "disparity map"—a special image that tells the robot how far away every point is.
Spotting the Circle: The tip of the needle has a special round marker. The robot uses a technique called Hough gradient transform to find this circle in the image, pinpointing its exact location.

By combining these steps, the robot can track the needle tip in real-time, updating its position every second (specifically, every 1000 milliseconds).

The Brain: The "Fuzzy" Controller

Once the robot sees where the needle is, it needs to move it there. This is where the Adaptive Fuzzy PID comes in.

To understand this, imagine you are trying to park a car in a tight spot.

A standard PID controller is like a driver who only reacts to how far off they are. If they are too far left, they turn right. But they might turn too hard, swing too far right, and then have to correct again, causing a shaky, jerky stop.
The Adaptive Fuzzy PID is like a driver who not only sees how far off they are but also how fast they are moving toward the spot. If they are moving fast and getting close, the driver gently eases off the gas before they overshoot.

The "Fuzzy" part is the secret sauce. In real life, things aren't always black and white (0 or 1). "Fuzzy logic" allows the robot to use rules like "If the error is small and getting faster, then slow down a little." The robot uses these rules to constantly tweak its own settings (the P, I, and D values) on the fly. It's like the robot is learning to drive the needle in real-time, adjusting its grip based on how the tissue feels and how the needle is moving.

The Results: Smoother and More Accurate

The researchers tested their new system in two ways: computer simulations and real-world experiments with a physical robot.

In the simulations:
When they compared their new "Fuzzy" brain against a traditional "Standard" brain, the results were clear. The new system was much calmer. It reduced the "overshoot" (swinging past the target) by about 35% and stopped shaking about 20% faster. It reached the target smoothly, without the jerky back-and-forth motion of the old system.

In the real experiments:
The team built a robot with a needle and tried to move it to specific targets.

Before fixing the errors: The robot was okay, but not perfect. The average error was about 2.4 mm on the X-axis, 3.2 mm on the Y-axis (the main direction the needle moves), and 1.8 mm on the Z-axis. The biggest mistake was 8.6 mm, which is too big for a delicate surgery.
The Fix: The researchers noticed a pattern in the mistakes. They realized the robot's motors were sometimes missing a tiny step (like a runner tripping over their own shoelace). They created a mathematical "compensation" to fix this.
After the fix: The results improved dramatically. The average error on the main feeding axis (Y-axis) dropped to 1.2 mm, and the biggest mistake was reduced to 3.4 mm.

Why This Matters

The paper concludes that this system successfully bridges the gap between "seeing" and "doing." By combining the robot's 3D vision with a smart, self-adjusting control system, they created a closed loop where the robot sees the needle, calculates the error, and instantly corrects its movement.

While the initial errors were a bit high, the final compensated results met the strict requirements for clinical surgery, which generally demand an accuracy of less than 2 mm. The study suggests that this approach—using two cameras to see and fuzzy logic to think—could be a reliable way to make future breast cancer biopsies safer, more precise, and less stressful for patients. It proves that with the right mix of vision and smart control, robots can learn to handle the delicate task of navigating soft human tissue.

Technical Summary: Adaptive Fuzzy PID Control of Breast Intervention Robot Based on Visual Detection

Problem Statement
The paper addresses the critical challenge of achieving high-precision control in breast intervention robots, specifically for tumor biopsy procedures. While visual detection technologies (such as MRI and ultrasound) are widely used for diagnosis, their integration into minimally invasive interventional robots for real-time needle guidance remains underdeveloped. Existing solutions often lack mature products capable of handling the nonlinearities of wire transmission and the uncertainties of soft tissue interaction. Furthermore, core control algorithms for high-precision operations often fail to meet the stringent clinical requirements for safety and accuracy, particularly in compensating for positional deviations during needle insertion.

Methodology
The authors propose a closed-loop control framework that integrates real-time binocular stereo vision with an adaptive fuzzy PID controller. The methodology is structured into three primary components:

Visual Perception and Needle Tip Localization:
- Hardware: A breast intervention robot with three degrees of freedom (X, Y, Z axes) is equipped with a binocular stereo camera system.
- Image Processing Pipeline: The system employs a workflow involving image preprocessing (grayscale conversion and median filtering), Bouguet stereo rectification to align image planes, and Semi-Global Block Matching (SGBM) for stereo matching to generate disparity maps.
- 3D Coordinate Extraction: To locate the needle tip, the system uses Hough gradient transformation to detect a spherical marker on the needle tip in the rectified left-eye image. By combining the pixel coordinates with the disparity map and camera intrinsic/extrinsic parameters, the system calculates the 3D spatial coordinates of the needle tip using triangulation principles.
Adaptive Fuzzy PID Control:
- Control Strategy: Instead of a fixed PID controller, the system utilizes an adaptive fuzzy PID controller. This controller takes the position error ( $e$ ) and the rate of change of error ($ec$) as inputs.
- Fuzzy Logic: A fuzzy inference engine with a dual-input, triple-output mechanism ( $e$ , $ec$ $\rightarrow$ $\Delta K_p$ , $\Delta K_i$ , $\Delta K_d$ ) is designed. It employs a rule base derived from expert knowledge to tune the PID parameters online.
- Implementation: The fuzzy logic uses seven linguistic variables (Negative Big to Positive Big) with triangular membership functions. The defuzzification is performed using the barycenter method to ensure smooth output adjustments.
Experimental Validation:
- Experiments were conducted using a hardware platform comprising the robot, an STM32F103VET6 microcontroller, and a DAUL-100M-60 binocular camera.
- The system was tested over an 8000ms period with 1000ms sampling intervals.
- An error compensation mechanism was applied to the Y-axis (the primary feed direction) based on curve fitting of pulse counts versus displacement to address step-out errors caused by motor resonance.

Key Contributions
The paper identifies three main contributions:

Real-time Vision Workflow: Development of a robust visual inspection workflow integrating Bouguet correction, SGBM matching, and Hough circle detection, achieving consistent extraction of needle tip 3D coordinates in static environments.
Adaptive Control Design: The design of an adaptive fuzzy PID controller with a dual-input, triple-output inference mechanism. This allows for online tuning of control parameters based on real-time error dynamics, addressing system nonlinearities without requiring a precise mathematical model of the soft tissue interaction.
Experimental Verification: Successful validation of the integrated system, demonstrating that error compensation can bring the average targeting error of the primary feed axis within clinical requirements.

Results

Simulation: Simulink simulations comparing the proposed adaptive fuzzy PID controller against a traditional PID controller showed that the proposed method reduced overshoot by approximately 35% and shortened settling time by about 20%.
Visual Detection: The system successfully generated stable disparity maps and extracted 3D coordinates within the 1000–4000 ms sampling window.
Motion Control:
- Initial Performance: Without compensation, the average errors were 2.4 mm (X-axis), 3.2 mm (Y-axis), and 1.8 mm (Z-axis). The maximum error on the Y-axis reached 8.6 mm.
- Compensated Performance: After applying pulse compensation to the Y-axis, the average targeting error was reduced to 1.2 mm, with a maximum error of 3.4 mm.
- Clinical Relevance: The compensated Y-axis average error meets the clinical accuracy requirement of less than 2 mm, although the maximum error (3.4 mm) exceeds this threshold.

Significance and Claims
The authors claim that this research bridges the technical gap between visual perception and adaptive control in breast intervention robots. By integrating interpretable fuzzy rules with real-time visual feedback, the system provides a deployable solution that does not rely on large-scale clinical datasets or precise dynamic models of soft tissue. The study positions this approach as a step toward autonomous and precise breast cancer diagnosis and treatment, offering a technical pathway for high-safety medical robots that balance interpretability, reliability, and engineering feasibility. The authors acknowledge that while the system meets average clinical accuracy standards, challenges remain in balancing rule base completeness with computational complexity and compensating for inherent visual feedback delays.