During the project, we foresee three journal publications to be generated, with a similar output of conference presentations expected. The three publications will be related to the three main working packages. The aim is to submit the two of them before the end of the project and complete the third one, right afterwards.
Combination of biomechanical modeling and learning
The first publication will cover the combination of biomechanical model and learning. The target is to publish this work in an engineering/biomechanics journal. For this publication, we are planning an experiment where we define three motoric tasks derived from a relevant rehabilitation task. There will be five volunteers participating in this study, and they will be asked to perform multiple repetitions of these tasks. During this, their respective EMG signals and kinematics will be measured and used for training the prediction algorithm.
To define how well the intention of motion is calculated, we perform a cross-validation, where will use the measurements of four volunteers to train the algorithm and see how well it can predict the motion of the fifth volunteer. By rotating the volunteers that we use for training and the volunteer we use for validation, we can quantify the ability of the algorithm to predict intention of new volunteers. We can also perform a cross-validation for all the trials of one volunteer, calculating like this how many repetitions are needed for each volunteer in order for the algorithm to predict accurately enough.
The results that will be presented are how good is the match between the actual kinematics of the volunteer and the prediction of the algorithm. Besides the fit of the measured vs predicted kinematics, we can also do an investigation of how long can the prediction horizon be valid, and how often should we recalculate the intention.
For this publication we should define the experimental protocol around month 8 of the project (December 2019) and the experiments should be performed on month 10 of the project (February 2019). The draft of the publication should be prepared by month 13 (May 2019).
Planing of the force profile
The second publication will cover the second work package about planning of the force profile of the robot. Once we have identified the force profile that the robot should apply on the human (through the prediction of the intention of motion), we should plan the trajectory of the robot that will allow it to apply this force profile in the most optimal way. The target is to publish this work in an engineering/robotics journal.
To obtain results for this publication, we will perform the following two experiments. Initially, we will define a specific force profile that the robot should follow. This profile should be within the limits of forces that are expected during rehabilitation tasks so that it remains relevant. Then the ability of the robot to apply the desired force profile will be quantified by measuring the applied force and comparing it to the setpoint. The same experiment can be also repeated for a rehabilitation relevant task and a volunteer. The volunteer can be asked to perform a specific task and a force profile will be calculated using the algorithm of work package 1. The desired force profile will be once again compared against the actual force (as measured with a load cell attached on the interface between the robot and the volunteer).
This experiment is related to the experiments of the third publication, therefore the protocol should be defined for both of them together. The protocol should be ready by month 14 of the project (July 2019), while the experiments are planned for month 21 of the project (January-February 2020). The first draft of the publication should be ready before the end fo the project (April 2020)
Results of the case study
The final publication will present the results of the clinical case study. It will be very similar to the study for the second publication, except that for this publication we will study actual patients. This will help us prove that this rehabilitation scheme can be applied on patients as well, not only healthy volunteers. It will also help us identify differences in the EMG or kinematics collection, and/or in the intention prediction between health and non-healthy volunteers. In this respect, the experiments will only involve tests with rehabilitation tasks.
We will use the same protocol as with the previous publication, so that the results can be compared. We will also have a recruiting period starting on month 14 of the project (July 2019), to ensure that we will have enough volunteers participating. The measurements will be performed at the end of the project (January-February 2020), while the publication is expected to be ready for submission after the project end.
