Transhumeral Arm Reaching Motion Prediction through Deep Reinforcement Learning-Based Synthetic Motion Cloning
Round 1
Reviewer 1 Report
In the presented study, the authors collected real motion data from six individuals performing multi-directional arm reaching tasks in the horizontal plane. Then, they generated synthetic motion data that mimicked similar arm reaching tasks by utilizing a physics simulation and DRL-based arm manipulation. The authors then trained a CNN-LSTM network with different configurations of training motion data, including DRL, real, and hybrid data sets, to test the efficacy of the cloned motion data. The results of evaluation showcased the effectiveness of the cloned motion data in training the ANN to predict natural elbow motion accurately across multiple subjects. Furthermore, motion data augmentation through combining real and cloned motion data sets demonstrated enhanced robustness of the ANN by supplementing and diversifying the limited training data. The authors proposed using the findings for creating synthetic data set resources for various arm movements and fostering strategies for automatized prosthetic elbow motion.
The study looks very good and promising. The authors should present further details of the training and testing phases of the DRL and CNN. The authors should also compare the prediction accuracy of the proposed models with the state-of-the-art models for proving the efficacy of the proposed models. The authors should specify the future direction of their research.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
This manuscript discusses the synthetic data generation issues used in artificial neural network for biomedical robotics applications. The whole paper is well written and the structure is well presented with clear illustrations. Here are few comments for authors.
1. Are the trained models still applicable and accurate enough when the elbow motion is not circular anymore, such as arbitrary hand swing motion or abrupt movements (e.g, higher speed)? The limitations of the used method should be mentioned for clarity.
2. Is there a fair quantitative analysis showing how much the efficiency can be improved by using the proposed synergetic data method? How much synergetic data and real data (and its percentage) is required to obtain a similar model or joint angle accuracy comparing to the case of using all real data for ANN training?
Readable
Author Response
Please see the attachment.
Author Response File: Author Response.pdf