Autonomous 6-DOF Manipulator Operation for Moving Target by a Capture and Placement Control System

Xiang Chen, Peilin Liu, Rendong Ying, Fei WenSensors 2022 · OpenReview

TL;DR

This paper presents a novel control system for a 6-DOF robotic manipulator to autonomously capture and place moving targets on a pipeline. The system combines an efficient online capture and offline placement strategy with an improved Position-Based Visual Servoing (PBVS) controller that leverages estimated velocity feedforward and a refined PID controller with dead zone and fuzzy tuning for enhanced speed, stability, and robustness.

What Problem Does This Paper Address?

The autonomous operation of robotic manipulators for tasks like capturing and placing moving objects, especially in industrial settings, presents significant challenges. These include accurately tracking dynamic targets, achieving precise and stable manipulation, and ensuring real-time performance. Existing visual servo control systems often struggle with issues like camera calibration inaccuracies, image noise, and the inherent nonlinearities and uncertainties of manipulator dynamics. Specifically, achieving fast, stable, and robust capture of moving targets while maintaining high operational efficiency is a key hurdle.

Methods

The authors propose a comprehensive system with several key components:

  • Online Capture and Offline Placement Strategy: To improve overall efficiency, the system employs an online strategy for capturing the moving target based on real-time visual feedback. For the placement phase, a fixed target pose allows for offline planning of the manipulator's trajectory, reducing computational load during operation.
  • Position-Based Visual Servoing (PBVS) Controller: This forms the core of the capture control. It's an enhanced PBVS controller that integrates:
  • Estimated Velocity Feedforward: A motion estimation method, combining Kalman filtering and interpolation, is used to accurately predict the target's pose and velocity. This predicted velocity is then used as a feedforward term to improve the system's response speed.
  • Refined PID Controller: The PID controller is modified in two ways:
  • Dead Zone Constraint: A dead zone is introduced to prevent excessive manipulator jitter caused by frequent small adjustments, particularly when the end-effector is close to the target.
  • Fuzzy Control Tuning: The PID coefficients (Kp, Ki, Kd) are adaptively tuned online using a fuzzy logic controller. This allows the system to dynamically adjust to noise, disturbances, and variations in the manipulator's dynamics, enhancing robustness.
  • Motion Estimation Method: This method uses a Kalman filter to predict the target's state (pose, velocity, acceleration). When the machine vision system's update rate is low, an interpolation technique is employed to bridge the gaps, ensuring a smooth and continuous estimation of the target's motion.

Datasets

The paper does not explicitly mention the use of specific public datasets for training or evaluation. Instead, the experiments were conducted on a custom-built ROS-Gazebo simulation platform, which models a 6-DOF manipulator, an eye-in-hand camera, a vacuum gripper, and a moving object on a pipeline. The performance was evaluated through simulations and described in terms of tracking errors and time spent in various operational phases.

Key Insights and Ideas

  • The synergy of online capture and offline placement is crucial for efficiency: By offloading placement calculations, the system can focus computational resources on real-time tracking and capture, leading to faster overall operation.
  • Velocity feedforward is essential for fast response: Directly incorporating estimated target velocity into the control loop significantly improves the manipulator's ability to keep pace with a moving object.
  • PID refinement is key to stability and robustness: The dead zone constraint effectively mitigates manipulator jitter, while fuzzy tuning allows the PID controller to adapt to dynamic environments and disturbances, outperforming standard PID and even feedforward-PID controllers.
  • Kalman filtering with interpolation provides robust motion estimation: This approach effectively handles noisy sensor data and variable update rates, providing reliable velocity estimates for feedforward control.

Key Results

  • Improved Dynamic Response: The developed controller demonstrated faster response times in reaching stable tracking stages compared to conventional PID, Velocity Feedforward PID (VFPID), and Fuzzy PID (FPID) controllers, both for low-speed (0.6 m/s) and high-speed (1 m/s) moving targets. For instance, in low-speed tracking, the developed controller achieved stable tracking in approximately 2.852 seconds, significantly less than PID (8.324 s) and FPID (4.930 s).
  • Enhanced Robustness: The system showed better performance under Gaussian noise and external joint disturbances. The error variation amplitude and adjustment time were significantly reduced compared to other controllers. For example, under a -0.15 radian joint disturbance, the developed controller had an error variation amplitude of 0.244 m along the x-axis, compared to 0.615 m for PID.
  • Reduced Overall Operation Time: In autonomous operation experiments, the developed system, utilizing the capture and placement strategy, achieved the shortest overall operation time (6.516 s) compared to PID (11.260 s), VFPID (8.347 s), and FPID (7.129 s).

Limitations and Future Work

  • Simulation-based validation: While the ROS-Gazebo simulation platform provides a robust testing environment, the authors acknowledge that real-world experimental validation is the next crucial step.
  • 3D space operation: The current work focuses on a pipeline scenario. Future work aims to extend the autonomous operation to more complex 3D environments.
  • End-to-end visual servo manipulation: Further research will explore a more integrated end-to-end visual servo manipulation system to further enhance robustness.

Summary updated

BibTeX
@inproceedings{chen2022autonomous,
  author = {Xiang Chen and Peilin Liu and Rendong Ying and Fei Wen},
  title = {{Autonomous 6-DOF Manipulator Operation for Moving Target by a Capture and Placement Control System}},
  booktitle = {Sensors},
  year = {2022},
  url = {https://openreview.net/forum?id=Hb61wWFSItw},
}
Figure