HomeEducationDoctorate (PhD & EngD)For current candidatesPhD infoUpcoming public defencesPhD Defence Jule Bessler | Safety first in rehabilitation robots! - Investigating how safety-related physical human-robot interaction can be assessed

PhD Defence Jule Bessler | Safety first in rehabilitation robots! - Investigating how safety-related physical human-robot interaction can be assessed

Safety first in rehabilitation robots! - Investigating how safety-related physical human-robot interaction can be assessed

The PhD Defence of Jule Bessler will take place in the Waaier building of the University of Twente and can be followed by a live stream.
Live stream

Jule Bessler is a PhD student in the department Biomedical Signals and Systems. (Co-)supervisors are prof.dr. J.H. Buurke from the faculty of Electrical Engineering, Mathematics and Computer Science and dr. G.B. Prange from Roessingh Research and Development.

In recent years, various robotic devices have been developed to be used in rehabilitation, assist patients, compensate for or alleviate disabilities. Those rehabilitation robots interact very closely with humans and transfer energy to their body to fulfil their purpose. This naturally introduces risks which have to be assessed carefully as rehabilitation robot use should be safe for patients and healthcare professionals. How can the safety of rehabilitation robots be ensured? This question needs to be answered for each device under development before it can go to market and be implemented in clinical practice. Safety certification is perceived as difficult for many rehabilitation robot developers. To shed light on safety challenges in rehabilitation robotics and take first steps towards addressing those challenges, the research presented in this thesis focused on the following research questions (Chapter 1):



1.       What are the most pressing risks and safety issues in rehabilitation robotics?

2.       How can those safety issues be avoided or managed?

3.       What are suitable methods to quantify metrics relevant for the identified safety issues?

4.       What are acceptable limit values to guarantee the safety of rehabilitation robot use?

In Chapter 2, the question on the most pressing safety issues was approached by means of a systematic literature review. We analyzed the occurrence and type of adverse events during the use of stationary robotic gait trainers to cover one common type of rehabilitation robots. The review included 50 studies involving 985 subjects. We noticed that information about adverse events is often incomplete or lacking completely. We found that soft tissue-related injuries and musculoskeletal complaints were the most commonly reported adverse events. Further analyses (Chapter 3) revealed that those are also the most frequently reported issues in other devices and that injuries in rehabilitation robot use can often be attributed to excessive forces applied to the human body by the device. More specifically, excessive loads on the soft tissue level and on the musculoskeletal level were identified as the most relevant hazards in physical interaction between humans and rehabilitation robots. We further identified important factors which can potentially lead to excessive force application, such as exceeding the anatomical range of motion and misalignments between the device and the anatomical structures of the user.

There can be many technical approaches to manage or avoid a certain safety issue and the solutions are often very individual and dependent on the device. Nevertheless, we joined forces with four other research and technology organizations within the EU-funded COVR project to approach safety on a system level and develop guidelines which are universally applicable across various domains dealing with robots that closely interact with humans. The approach is described in detail in Chapter 3. Next to following the applicable directives and regulations (for rehabilitation robotics, the Medical Device Regulation applies) and implementing procedures for risk management laid down in standards, (rehabilitation) robot developers can perform validation experiments. Instead of focusing on the exact technical implementation of a risk reduction measure, those validation experiments can be designed in such a way that a certain behavior is tested on the system level. We combined the knowledge about adverse event occurrence and risk factors with experience regarding safety validation experiments in other domains to develop step-by-step procedures for testing the safety of rehabilitation robots. During this process, it became apparent, that there is a lack of knowledge on measurement techniques and devices to assess the forces of interest and on safe limit values for those forces.

To fill some of the knowledge gaps regarding testing procedures, a prototype measuring device was developed to assess the force interplay between a human arm and a splint. The proof of concept study presented in Chapter 4 describes the measuring device based on force sensitive resistors and a force/torque sensor and discusses some drawbacks and potential improvement options. Force sensitive resistors are fairly cheap and easy to use but issues such as hysteresis and the loss of information when forces are transmitted between the sensing areas of the sensors limit their reliability when used at the contact interface of rehabilitation robots. Load cells are reliable and can measure forces in all directions, but are not slim enough to be used at the skin-robot interface. To accurately investigate the safety-relevant force interplay at the human-robot interface on the soft tissue level, thin and flexible sensors are needed. Those sensors should be able to record normal and shear stresses and should not alter the interface.

A direct assessment of loads on the musculoskeletal system in vivo is almost impossible. Chapter 5 and Chapter 6 therefore focused on ways to assess these loads with an instrumented leg simulator (dummy), specifically focusing on the risk factor misalignment. The leg simulator developed for this study was equipped with a 6 degrees of freedom force and torque sensor to assess the loads on its knee joint. We attached a passive knee brace to the leg simulator and imposed misalignments (internal/external rotation, proximal/distal translation, anteroposterior translation) to assess their effects on joint load during swing. In Chapter 5, those experiments were performed with the original leg simulator, equipped with a polyurethane foam to mimic soft tissue. In Chapter 6, we repeated the experiments (excluding anteroposterior translational misalignment) with 3 additional materials to mimic soft tissue, namely one softer foam and two silicone/gel materials which were harder than the original foam and had been suggested as suitable materials to mimic soft tissue behavior in literature. We found that misalignments of a leg exoskeleton in each of the investigated directions can increase internal knee forces and torques during swing to a multiple of those experienced in a well-aligned situation. The investigations further showed that those changes in joint loads induced by misalignments are significantly affected by the characteristics of (artificial) soft tissue. There was a tendency for stronger increases of peak joint loads in harder materials. This research showed that dummies can be a suitable tool to measure joint loads. Future research should work towards improving the accuracy with which dummies simulate the physical human-robot interaction.

The study presented in Chapter 7 focused on investigating how loads applied by rehabilitation robotics affect comfort and safety. More specifically, we investigated how discomfort develops and changes over time when forces are exerted repetitively and for extended durations through a rigid cuff. The experimental cuff and the exerted force patterns were based on what is known about the interaction with gait exoskeletons. Three force patterns were applied to healthy participants (n=15) of two age groups, who continuously indicated their perceived (dis-)comfort on a visual analog scale during each force pattern. A baseline trial was used to detect each participant’s individual discomfort detection threshold. The two other trials consisted of repetitive forces applied for 30 minutes (normal force only and normal and shear force combined) and the magnitudes were based on the individual discomfort detection threshold. The discomfort detection thresholds (median 100 N; range 40 N to >230 N) and times to discomfort detection in repetitive force trials (median 4.1 minutes in repetitive normal force and 5.4 minutes in repetitive normal and shear force) varied strongly between participants. The force applications resulted in skin reddening but usually no pain. No relationships between discomfort and age group, force pattern, environmental conditions or other participant characteristics were found. There were no significant differences between the two repetitive force patterns. We saw that discomfort increases over time when a force that is perceived as slightly uncomfortable in one single contact is applied repetitively.

Finally, in Chapter 8, the main findings of this thesis are discussed, giving rise to recommendations for future research and development concerning mechanical safety of rehabilitation robots. In addition, implications for clinical practice and (guidelines for) safety assessments are presented. While this research has taken steps in several directions to close knowledge gaps and simplify safety assessments in rehabilitation robots, a number of issues remain. We found that excessive loads on the soft tissue and musculoskeletal tissue can be considered the most relevant hazards in physical interaction between rehabilitation robots and their users. The nature of interaction in rehabilitation robotics, characterized by continuous contacts, cyclic loading, vulnerable users, and sometimes uncontrolled environments, makes safety considerations complex. Even relatively small forces can lead to hazardous situations when they are e.g. applied for long durations, to impaired body structures, in interfaces with peak stresses or unfavorable microclimates. Safety validation experiments can be a useful approach to test physical human-robot interaction, preferably without a human in the loop, and develop mitigation strategies to reduce (peak) stresses and loads. We have discussed that additional research is required to accurately measure force distributions in all relevant directions at the contact interface between human and robot without affecting the force interplay. It has been established that misalignments are a prominent issue in exoskeleton use. While we have shown that misalignments can affect knee joint loads significantly in a dummy during swing, further research should investigate whether misalignment would lead to (potentially) harmful loads in weightbearing situations. Our research showed that discomfort increases over time when repetitive loads are applied through an exoskeleton cuff-like interface. Perception of comfort varies considerably between subjects. While it will be difficult to establish generally applicable guidelines for acceptable time-force relationships, future research can extend current knowledge by covering larger subject groups including patients, and other force patterns. The research in this thesis provided valuable insights into current safety issues and research gaps regarding rehabilitation robot safety. It is a first step towards building a knowledge base which can support the development and market entrance of safer rehabilitation robots though comprehensive guidelines.