Neuro-control with online adaptation for a Knee-Ankle-Foot-Orthosis
Georg-August-Universität Göttingen, 3. Physikalisches Institut, Germany
Bernstein Focus Neurotechnology Göttingen, Germany
Bernstein Center for Computational Neuroscience Göttingen, Germany
Otto Bock HealthCare GmbH, Germany
A Knee-Ankle-Foot-Orthosis (KAFO) is a modular lower-extremity orthosis prescribed to people with gait disability which might be, e.g., caused by diseases or injury to brain or spinal-cord. The KAFO should support, correct and assist the movement of the corresponding affected joints. Traditional KAFOs are restricted by a gait depending switch of the joints based on (electro-) mechanic non-adaptive switches. So common disturbances (floor unevenness, obstacles, ramps) cannot be mastered in a satisfactory way. Novel approaches include active elements into the orthosis, which do not directly act on the movement. Instead they adjust the compliance leading to new challenges for the controller of such actuators, which are difficult to handle with traditional approaches.
Thus new technologies have to be developed to improve control and to overcome the shortcomings of traditional non-adaptive approaches, thus solving the problem of efficient actuator control. Development of advanced orthotic devices is held back by the vast number of possible indications as well as by the wide range of neuromuscular variability within a specific patient group (Yakimovich et al., 2009). The development of advanced devices is therefore imposing the need for individual (online) adaptation of gait parameters to allow adaptation (1) to changing environments like slopes, stairs etc. as well as to gait parameters like stride length/frequency and (2) to the individual patients with respect to physiological conditions. To do so, we have employed a reflexive neuro-controller as inspired by RunBot (Manoonpong et al., 2007), embedded to a KAFO based on a controllable hydraulic damper, derived from OttoBock's C-Leg©.
Where traditional control approaches are designed with a fixed behaviour set, which is fit on the patient using an equally fixed set of parameters, the design of a neural controller allows extensive behaviour adaptation at run time, i.e. time continuous output signal modulation and modification, and smooth transitions between different time periodic output patterns. This improves the flexibility and thus the range of applications for the device. At the same time, the controller is implemented with fewer assumptions on the patient's conditions in mind and thus allows more patients to take advantage of it.
While these developments are advantageous, the patients desire for predictable and secure operation of the device are of utmost importance. In this study approaches to tackle these problems on the design level of the neural controller are presented, e.g., to achieve secure behaviour by reflexive fail safe mechanisms and predictability by restricting the possible degrees of freedom for behaviour adaptation. Thus allowing the application of unsupervised adaptation for medical devices.
This research was supported by the BMBF-funded BFNT Göttingen with grant number 01GQ0810 (project 3A) and BCCN Göttingen with grant number 01GQ1005A (project D1) and the Emmy Noether Program (DFG, MA4464/3-1).
Bernstein Conference 2012, Munich, Germany, 12 Sep - 14 Sep, 2012.
Data analysis, machine learning, neuroinformatics
(2012). Neuro-control with online adaptation for a Knee-Ankle-Foot-Orthosis.
Front. Comput. Neurosci.
Bernstein Conference 2012.
11 May 2012;
12 Sep 2012.
Mr. Jan-Matthias Braun, Georg-August-Universität Göttingen, 3. Physikalisches Institut, Göttingen, Niedersachsen, 37073, Germany, email@example.com