An adjusted dipole model for rectangular electromagnetic coils

Abstract

Magnetic actuation is an efficient way for remote wireless control of micro-robots. One key factor for efficient actuation is an accurate mapping of the magnetic field components and the field gradients that generated by the magnetic sources. Usually, magnetic dipole model or interpolation method is used to estimate the magnetic field. These methods are not well suited for the control of micro-robots by electromagnetic coils. In this paper, we present an adjusted magnetic dipole model for the rectangular electromagnetic coil. The proposed adjusted dipole model is based on the magnetic dipole model. This model can provide accurate estimation of the magnetic field and gradient, meanwhile simplifying the calculation. Experimental results show that the proposed model work well and better than the original magnetic dipole model.

More information

Full text and more information

Towards transferring skills to flexible surgical robots with programming by demonstration and reinforcement learning

Abstract:
Flexible manipulators such as tendon-driven ser-
pentine manipulators perform better than traditional rigid
ones in minimally invasive surgical tasks, including navigation
in confined space through key-hole like incisions. However, due
to the inherent nonlinearities and model uncertainties, motion
control of such manipulators becomes extremely challenging.
In this work, a hybrid framework combining Programming by
Demonstration (PbD) and reinforcement learning is proposed
to solve this problem. Gaussian Mixture Models (GMM),
Gaussian Mixture Regression (GMR) and linear regression are
used to learn the inverse kinematic model of the manipulator
from human demonstrations. The learned model is used as
nominal model to calculate the output end-effector trajectories
of the manipulator. Two surgical tasks are performed to
demonstrate the effectiveness of reinforcement learning: tube
insertion and circle following. Gaussian noise is introduced to
the standard model and the disturbed models are fed to the
manipulator to calculate the actuator input with respect to the
task specific end-effector trajectories. An expectation
maximization (E-M) based reinforcement learning algorithm is
used to update the disturbed model with returns from rollouts.
Simulation results have verified that the disturbed model can
be converged to the standard one and the tracking accuracy is
enhanced.
More Information:

Motion planning for flexible surgical robots

Project Goals

A constrained tendon-driven serpentine manipulator is developed in our previous work, high level intelligence is expected to make manipulator system working autonomously. Considering the limited and confined working space of surgical operation, this project is aim to develop novel motion planning techniques for our surgical tendon-driven serpentine manipulator, which is expected to assist surgeon operation more accurate and convenience.

Approaches

In the clinical environment, chances are high that the manipulator may bump into neighboring tissue and organs and cause additional damages. For medical manipulator working within human body, optimal and accurate trajectory planning is the key enabler of surgical security because of any additional damage to proper functioning organs is intolerable. Moreover, the surgery usually has many additional disturbances including breathe, physical hand tremor and tiny displacement of the organs. These uncertainties require the planning algorithm to have good robustness to avoid damaging the proper functioning organs. As the most important factor that can lower the risk of additional damages, less sweeping area of manipulator motion results in more disturbances tolerance capability. Especially when energy cost of different planned trajectories are in the same level, the less sweeping area of whole manipulator body becomes more attractive for physician and reduce unpredictable risks in transoral procedure applications. Therefore, we propose a three dimensional neural dynamic planning algorithm which introduces sweeping area as a very important factor in neural stimulation propagation.

The three dimensional neural network is show in Figure 1, each neuron is connected with its adjacent 26 neurons. In our minimum planning model, the start state with highest activity propagates stimulation to whole network through connective weight. On the other hand, the configuration parameters representing obstacle collision hold the lowest value and do not connect with neighboring neurons. As a result, the neural stimulation spread as water ran down from a height place in planning neural dynamic field building, which is shown in Figure 2. Simultaneously, the robot starts from target state and looks for the highest state as climbing mountain. When robot reach the start state configuration parameter, the planning is finished. Then, the final motion sequence will be obtained by reversing the planned trajectory, which is from start state to target state.

Fig.1 Three dimensional neural dynamic model

Fig.1 Three dimensional neural dynamic model

Fig.2 Neural dynamic field, which stimulation spread from start state to whole map.

Fig.2 Neural dynamic field, which stimulation spread from start state to whole map.

Results

At first, we test our planning algorithm in representative simulation scenarios and compare with other famous planning algorithms, such as traditional neural dynamic planning algorithm, RRT and RRT*.

(a) Minimum sweeping area planning algorithm

(a) Minimum sweeping area planning algorithm

(b) Neural dynamic algorithm

(b) Neural dynamic algorithm

(c) RRT*

(c) RRT*

Fig.3 Simple case: a comparison simulation on a simple scenario consisting of two obstacles in the middle of the map is used for first test. The tendon-driven serpentine manipulator is left bended at the beginning and expected to reach the upper right target point. The manipulator is presented by green color (free bending segments) and red color (constrained bending segments). The distal tip trajectory is presented by a dash line. The performance of different planning algorithms in terms of sweeping area and obstacle avoidance ability are shown. Particularly, the area of green parts can be seen as the sweeping area of manipulator approximately

(a) Minimum sweeping area planning algorithm

(a) Minimum sweeping area planning algorithm

(b) Neural dynamic algorithm

(b) Neural dynamic algorithm

(c) RRT*

(c) RRT*

Fig.4 Complex case: a comparison simulation on a complex scenario consisting of four obstacles are used for test. The
tendon-driven serpentine manipulator is left bended at the beginning and expected to reach the target point surrounded by three obstacles in the upper right area. The differences in sweeping area and obstacle avoidance ability are shown obviously among different algorithms guidance.

(a) Minimum sweeping area planning algorithm

(a) Minimum sweeping area planning algorithm

(b) Neural dynamic algorithm

(b) Neural dynamic algorithm

(c) RRT*

(c) RRT*

Fig.5 Tubular case 1: A virtual tubular clinical map in simulation is conducted in advance, where target is at the right side sub-branch. The tendon-driven serpentine manipulator is straight on the bottom of the tubular at the beginning.

(a) Minimum sweeping area planning algorithm

(a) Minimum sweeping area planning algorithm

(b) Neural dynamic algorithm

(b) Neural dynamic algorithm

(c) RRT*

(c) RRT*

Fig.6 Tubular case 2: The same virtual tubular clinical map as Tubular case 1, but target is at the left side sub-branch and start point at the right side sub-branch. The tendon-driven serpentine manipulator is expected to move from right sub-branch to left one as manipulator moving in clinical operation.

Moreover, experiments are conducted in environments built by Lego blocks, where obstacle configurations are similar to simulation cases. In this experiment stage, the tendon-driven serpentine manipulator is expected to execute same motion sequences that are generated from complex case, tubular case 1 and tubular case 2 in simulation studies. The experimental results in phantom test are shown in video 2 (Phantom test on Lego bricks).

Finally, a preliminary transoral trials on cadaver human head is conducted to at Khoo Teck Puat Advanced Surgery Training Centre (ASTC), National University of Hospital, Singapore. The panorama of cadaver transoral experiments can be found in video 3. The operations on compute and corresponding softwares are shown in video 4. The experimental data are ploted by MATLAB, and four algorithms comparisons are shown in video 5.

People Involved

Visiting PhD Student: Yanjie Chen
PhD Student: Wenjun Xu
Project Investigators: Hongliang Ren

Publications

[1] Yanjie Chen, Zheng Li, Wenjun Xu, Hang Zhong, Yaonan Wang and Hongliang Ren, “Minimum Sweeping Area Motion Planning for Flexible Serpentine Surgical Manipulator with Kinematic Constraints”, IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS 2015), Accepted.
[2] Yanjie Chen, Wenjun Xu, Zheng Li, Shuang Song, Chwee Ming Lim, Yaonan Wang, and Hongliang Ren, “Safety-Enhanced Motion Planning with Minimum Sweeping Area for Flexible Surgical Manipulators using Neural Dynamics”, IEEE Transactions on Cybernetics, Submitted.

Videos

-Support powerpoint.

-Phantom test on Lego bricks.

-The panorama of cadaveric transoral experiments.

-The compute vision and corresponding software during clinical experiments.

-The surgical robot shape motion in experiments.

 

Variable Selection Based on Information Tree for Spectroscopy Quantitative Analysis

Abstract

Spectroscopy is a fast and efficient component analysis method, and full spectrum prediction model may be redundant and inaccurate. This paper proposes a variable selection method based on information tree for spectroscopy quantitative analysis. Firstly, a feature training set that indicates the information of the selected variables is generated. Then, the partial least squares (PLS) is performed on the spectral calibration set, and root-mean-square error of cross-validation is used to evaluate the feature training set. According to the corresponding evaluation results, the information gain of each wavelength is calculated. The wavelength with maximum information gain is defined as the root node, and an information tree is built based on the information gain where each leaf node represents a wavelength. The final selection result is a conjunction path of the leaf nodes that has bigger information gain. The full spectrum PLS, the uninformative variable elimination with PLS method, the genetic algorithm with PLS method and the proposed method are conducted on the real spectral dataset of flue gas, and the effectiveness of the methods are compared and discussed. The experimental results verify that the prediction precision and the compression ability of the proposed method is higher.