As with any research project, there is an inherent level of risk regarding the foreseen tasks. This rises from the fact that several of these tasks have not been performed by other groups and therefore their successful completion is not guaranteed. Another factor of risk is that the allocated resource for each tasks might be under-(or over-)estimated. However, by identifying these potential risks, we can create a contingency plan that helps us cope with them if they manifest themselves.
We have identified four major risks for this project. In the following sections, we outline the risks and we provide a contingency and prevention measures.
On board planning is not feasible in real-time
Risk level: moderate
It is unknown how computationally intensive are the tasks that we are planning. Since there are several novelties in this project, we are not sure whether the computational power of a standard laptop will be enough to measure and analyse all the measured signals, calculate the intention and perform the planning of the robot. Needing to perform this in real-time increases the difficulty and the risk.
However, as a preventive measure we are planning to optimise the code so that it requires as little computational power as possible. In case the computational power will not be enough, the tasks are going to be run on dedicated computers. We have already a plan to acquire and process the measured signals (EMG and force) on a separate platform (compactRIO from National Instruments), and if needed, we can also offload the intention and planning calculations on the dedicated FPGA of the same platform. These FPGA processors can perform a great number of calculations in real-time and it is guaranteed that it will be able to perform the task in real-time.
Trajectory for all the tasks is difficult to predict
Risk level: moderate
The goal of the project is to ultimately be able to predict any kind of motion that a patient is trying to perform, with a relatively long prediction horizon. However, this goal might be too ambitious to be achieved within this project. The motions and the muscle activation patterns that generate them might be too variable, and therefore difficult to predict with such low training samples.
To avoid this risk, we initially aim at using robust learning algorithms that can deal with high uncertainty (e.g. Deep learning). Furthermore, we will start by training the learning algorithm to specific tasks, especially on tasks that are usually performed during a rehabilitation session. This way, the uncertainty will be reduced and will allow the algorithm to perform at least in some cases.
Robot arm cannot apply desired forces
Risk level: low
We have a Cyton Gamma robot at our disposal in the lab, which we will use for the initial tests of the planning. However, since this robot is limited in its capabilities (e.g. it is not possible to control the torque of the actuating motors), it will not be the robot that we will use for the final studies. Instead, we are planning to use an industrial robot (Baxter, from Rethink Robotics), which is available through a partner of the university (Accenture). This robot has been used for tasks collaborating with humans, therefore it is expected that it is able to deliver the payload required by this project.
Low number of participating patients
Risk level: low
To have relevant and reliable results for this project, we need a significant number of patients to participate in the validation study. The current aim is 10 volunteers, that ideally have similar pathology and fulfill certain criteria. It is possible that several patients might be reluctant to work with an innovative robotic arm due to fear of injury or pain during the study. To prevent this risk, we are planning to hold an info session for the patients, where we will outline the scope of the project, the study protocol and any potential dangers associated with it. We hope that this will make them more familiar with the developments of the project and therefore more keen to volunteer to participate.
However, in the worst case where we are not able to recruit the target number of volunteers, we will still perform the validation studies using healthy volunteers. This can still provide relevant results and validate for a healthy population the technical innovations of this project.
Safety of patients
Risk level: moderate
The safety of the volunteers participating in these studies is of outmost importance. Robots have been used in the past together with humans, however the development of new controllers should always be treated with extra care to avoid any potential injury of the volunteers. To reduce this risk, we aim at safeguarding both the software and the hardware that will be used during this project. At the software level, we are planning to include limiters for the speed and force that the robot can apply during its operation. This can create a safer environment around the robot, since it will not be able to perform any rapid motions that could injure a volunteer. Furthermore, on the hardware side, we are planning to keep the volunteers as away as possible from the robot. The volunteers will not be directly attached on the robotic arm, but be connected to the robot through a mechanical link that can get easily detached from the patients arm (possible connection just by holding an object attached on the robot). Finally, two hard stop switches will be present at a reachable position, so that the volunteer or the operator of the robot can immediately stop the robot in case something goes wrong.