An official website of the United States government
Significance: The performance of traditional approaches to decoding movement intent from electromyograms (EMGs) and other biological signals commonly degrade over time. Furthermore, conventional algorithms for training neural network-based decoders may not perform well outside the domain of the state transitions observed during training. The work presented in this paper mitigates both these problems, resulting in an approach that has the potential to substantially he quality of live of people with limb loss. Objective: This paper presents and evaluates the performance of four decoding methods for volitional movement intent from intramuscular EMG signals. Methods: The decoders are trained using dataset aggregation (DAgger) algorithm, in which the training data set is augmented during each training iteration based on the decoded estimates from previous iterations. Four competing decoding methods: polynomial Kalman filters (KFs), multilayer perceptron (MLP) networks, convolution neural networks (CNN), and Long-Short Term Memory (LSTM) networks, were developed. The performance of the four decoding methods was evaluated using EMG data sets recorded from two human volunteers with transradial amputation. Short-term analyses, in which the training and cross-validation data came from the same data set, and long-term analyses training and testing were done in different data sets, were performed. Results: Short-term analyses of the decoders demonstrated that CNN and MLP decoders performed significantly better than KF and LSTM decoders, showing an improvement of up to 60% in the normalized mean-square decoding error in cross-validation tests. Long-term analysis indicated that the CNN, MLP and LSTM decoders performed significantly better than KF-based decoder at most analyzed cases of temporal separations (0 to 150 days) between the acquisition of the training and testing data sets. Conclusion: The short-term and long-term performance of MLP and CNN-based decoders trained with DAgger, demonstrated their potential to provide more accurate and naturalistic control of prosthetic hands than alternate approaches.
more »
« less
Semi-Supervised Adaptive Learning for Decoding Movement Intent from Electromyograms
This paper presents an adaptive learning algorithm for predicting movement intent using electromyogram (EMG) signals and controlling a prosthetic arm. The adaptive decoder enables use of the prosthetic systems for long periods of time without the necessity to retrain them. The method of this paper employs a neural network-based decoder and we present a method to update its parameters during the operation phase. Initially, the decoder parameters are estimated during a training phase. During the normal operation, the parameters of the algorithm are updated in a semi-supervised manner based on a movement model. The results presented here, obtained from a single amputee subject, suggest that the approach of this paper improves long-term performance of the decoders over the current state-of-the-art with statistical significance.
more »
« less
- Award ID(s):
- 1533649
- PAR ID:
- 10121192
- Date Published:
- Journal Name:
- 27th European Signal Processing Conference, EUSIPCO 2019
- Volume:
- 1
- Issue:
- 1
- Page Range / eLocation ID:
- 86
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Many control methods have been proposed for powered prosthetic legs, ranging from finite state machines that switch between discrete phases of gait to unified controllers that have a continuous sense of phase. In particular, recent work has shown that a mechanical phase variable can parameterize the entire gait cycle for controlling a prosthetic leg during steady rhythmic locomotion. However, the unified approach does not provide voluntary control over non-rhythmic motions like stepping forward and back. In this paper we present a phasing algorithm that uses the amputee’s hip angle to control both rhythmic and non-rhythmic motion through two modes: 1) a piecewise (PW) function that provides users voluntary control over stance and swing in a piecewise manner, and 2) a unified function that continuously synchronizes the motion of the prosthetic leg with the amputee user at different walking speeds. The two phase variable approaches are compared in experiments with a powered knee-ankle prosthesis used by an above-knee amputee subject.more » « less
-
In this paper, we are the first to propose a new graph convolution-based decoder namely, Cascaded Graph Convolutional Attention Decoder (G-CASCADE), for 2D medical image segmentation. G-CASCADE progressively refines multi-stage feature maps generated by hierarchical transformer encoders with an efficient graph convolution block. The encoder utilizes the self-attention mechanism to capture long-range dependencies, while the decoder refines the feature maps preserving long-range information due to the global receptive fields of the graph convolution block. Rigorous evaluations of our decoder with multiple transformer encoders on five medical image segmentation tasks (i.e., Abdomen organs, Cardiac organs, Polyp lesions, Skin lesions, and Retinal vessels) show that our model outperforms other state-of-the-art (SOTA) methods. We also demonstrate that our decoder achieves better DICE scores than the SOTA CASCADE decoder with 80.8% fewer parameters and 82.3% fewer FLOPs. Our decoder can easily be used with other hierarchical encoders for general-purpose semantic and medical image segmentation tasks.more » « less
-
This paper presents a voltage control design algorithm for a three-phase voltage source converter (VSC) connected to a linear load using a passive LCL filter. The controller is based on an optimal regulator, combined with the integral of the voltage error, which is called LQI that achieves null tracking of the error. The main objective of this paper is to present an algorithm to design a voltage controller through frequency analysis of the singular values of the system, the weight of the states involved in the system, the movement of the closed poles and their respective step response to evaluate performance and robustness against load changes. The simulation results show a satisfactory operation of the voltage controller with fast recovery after a resistive load changemore » « less
-
This paper proposes an extremum seeking controller (ESC) for simultaneously tuning the feedback control gains of a knee-ankle powered prosthetic leg using continuous-phase controllers. Previously, the proportional gains of the continuous-phase controller for each joint were tuned manually by trial-and-error, which required several iterations to achieve a balance between the prosthetic leg tracking error performance and the user's comfort. In this paper, a convex objective function is developed, which incorporates these two goals. We present a theoretical analysis demonstrating that the quasi-steady-state value of the objective function is independent of the controller damping gains. Furthermore, we prove the stability of error dynamics of continuous-phase controlled powered prosthetic leg along with ESC dynamics using averaging and singular perturbation tools. The developed cost function is then minimized by ESC in real-time to simultaneously tune the proportional gains of the knee and ankle joints. The optimum of the objective function shifts at different walking speeds, and our algorithm is suitably fast to track these changes, providing real-time adaptation for different walking conditions. Benchtop and walking experiments verify the effectiveness of the proposed ESC across various walking speeds.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
