Human-Manipulator Interface Using Particle Filter

Guanglong Du, Ping Zhang, Xueqian WangThe Scientific World JOURNAL 2014 · DOI

TL;DR

This paper presents a novel human-robot interface system that uses a 3D camera (Kinect) and a particle filter (PF) to track human hand movements for robot teleoperation. To overcome limitations in precision and human motor control, they introduce an Adaptive Multispace Transformation (AMT) method, which significantly improves accuracy and reduces operation time compared to existing methods.

What Problem Does This Paper Address?

Controlling robots remotely, especially for tasks requiring high precision, is challenging. Traditional methods often involve unnatural hand movements or require cumbersome physical interfaces. Vision-based approaches offer a more natural interaction, but existing markerless tracking methods struggle with accuracy, especially during brief sensor occlusions or when dealing with inherent human limitations like motor tremors and perceptual inaccuracies. This makes it difficult for operators to perform delicate manipulation tasks precisely and efficiently.

Methods

The core of the system involves:

  • Human Hand Tracking: A 3D camera (Kinect) captures color and depth images. The Camshift algorithm is used for initial marker tracking on a glove, leveraging color probability distribution in HSV space for robustness.
  • Position Estimation: A Particle Filter (PF) is employed to estimate the hand's translation. The PF uses a set of weighted samples (particles) to represent the probability distribution of the hand's state, allowing for robust estimation even with noisy sensor data.
  • Adaptive Multispace Transformation (AMT): This is the paper's key innovation. AMT addresses the limitations of human perception and motor control. It involves two scaling processes: one scales the human operator's movements, and the other scales the robot's end-effector movement. This allows for more precise control, mapping large human movements to fine robot adjustments and vice-versa, improving accuracy and reliability. The system also incorporates a velocity control mechanism to mitigate data jitter caused by hand tremors.

Datasets

Not discussed in the paper. The experiments were conducted in a lab environment using a GOOGOL GRB3016 robot and a custom setup with a 3D camera and a target object.

Key Insights and Ideas

  • AMT for Human-Robot Precision: The AMT method is crucial for bridging the gap between human limitations (tremors, perceptual errors) and the need for high-precision robot manipulation. It effectively filters and scales movements to enhance operator control.
  • PF for Robust Tracking: While PF is a standard technique, its application here for hand translation estimation, combined with Camshift for initial tracking, provides a robust foundation for the system, especially in handling potential sensor dropouts.
  • Velocity Control for Stability: Using velocity control, rather than direct position control, helps to smooth out the control signals and mitigate the effects of natural human hand tremors, leading to more stable robot operation.
  • Natural Interaction: The system enables a more intuitive and natural way for humans to control robots, allowing them to focus on the task rather than on complex command sequences.

Key Results

The paper demonstrates that their proposed system, incorporating PF and AMT, significantly outperforms a previous vision-based method ([6]) in terms of:

  • Accuracy: The mean translation errors for their method were 3.02 mm (X) and 3.20 mm (Y), with a mean rotation error of 0.91 deg (Z). Method [6] had significantly higher errors: 5.77 mm (X), 4.54 mm (Y), and 2.44 deg (Z). The rotation errors of their method were less than half of method [6].
  • Efficiency: The average operation time for their method was 131.0 seconds, compared to 165.66 seconds for method [6]. This speedup is attributed to the AMT's ability to allow larger robot movements from smaller human inputs, reducing the need for frequent corrections.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that human operators still face perceptual and motor limitations, making extremely high-precision tasks challenging even with the AMT.
  • The system relies on markers on a glove, which, while less intrusive than full body markers, still requires the operator to wear specific equipment.
  • The current system does not include force feedback, which would be beneficial for tasks requiring delicate contact or assembly.
  • Future Work:
    • The authors suggest merging gesture control with voice control for more flexible and advantageous robot operation.
  • Future research could explore incorporating force feedback for more nuanced manipulation tasks.
  • Further development could aim to reduce reliance on markers or explore fully markerless approaches.

Summary updated

Abstract

This paper utilizes a human-robot interface system which incorporates particle filter (PF) and adaptive multispace transformation (AMT) to track the pose of the human hand for controlling the robot manipulator. This system employs a 3D camera (Kinect) to determine the orientation and the translation of the human hand. We use Camshift algorithm to track the hand. PF is used to estimate the translation of the human hand. Although a PF is used for estimating the translation, the translation error increases in a short period of time when the sensors fail to detect the hand motion. Therefore, a methodology to correct the translation error is required. What is more, to be subject to the perceptive limitations and the motor limitations, human operator is hard to carry out the high precision operation. This paper proposes an adaptive multispace transformation (AMT) method to assist the operator to improve the accuracy and reliability in determining the pose of the robot. The human-robot interface system was experimentally tested in a lab environment, and the results indicate that such a system can successfully control a robot manipulator.
BibTeX
@article{du2014human,
  author = {Guanglong Du and Ping Zhang and Xueqian Wang},
  title = {{Human-Manipulator Interface Using Particle Filter}},
  journal = {The Scientific World JOURNAL},
  year = {2014},
  doi = {10.1155/2014/692165},
}
Figure