This blog post takes a few relevant observations, and assumptions, throws them up in the air and sees if they turn into sunshine.
- If anything has brought us forward, it is also the ability to find relevant short cuts. We do not always have to invent the wheel when really we just want a variation of it.
- If there is any acutal success story where academic research was required to leverage consumer market for a prosthetic limb, it is that of Otto Bock’s C-leg.
- If we can understand what the concepts are for getting a C-leg successfuly built, marketed and sold, we should be able to take generalized aspects of it to formulate success elements for prosthetic hands, grippers or arms.
Background
While the idea of a microprocessor controlled knee was created earlier [link], no marketable solution was available in due course. “In the early 1990s, Kelly James, an engineer at the University of Alberta, Edmonton, Canada, developed the C-Leg, the first leg with microprocessor-controlled swing and stance phases. Buying the rights from the university, he traveled around the world to interest prosthetic manufacturers in his invention (“A Leg Up,” by Isabelle Gallant, U of A Engineer, Spring 2011). However, he didn’t receive any commercial interest until German manufacturer Ottobock bought the patent in 1992 and launched the groundbreaking technology.”.
Then, based on work betweeen 1995 and 1998, a doctoral thesis at the ETH Zurich described an intelligently, microprocessor controlled knee for above knee prostheses built from available and affordable materials [1].
That research was performed 1995 to 1998, financially supported by Otto Bock, and Otto Bock presented its first C-Leg in 1997.
The rest is history. If ever there was a leap in performance of prosthetic function, ever, it was the C-Leg. No prosthetic hand ever came close to achieving this level of success.
So this particular doctoral thesis seems to contain some possibly interesting ingredients worthwhile looking at. As any doctoral thesis here is public record, and a copy of it must be made available at the public library, I borrowed a copy for further information.
There are some other prosthetic developments, however, nowhere else is academic research anywhere near that successful as in the instance of the C-leg:
- Otto Bock Michelangelo hand; the mechanism seems to come from American DARPA or other army research and probably was just built, the first glove was a great design work. So there is no analytical approach comparable to the C-Leg. It is too heavy, it does not work with prosthetic gloves really, it is not sturdy.
- i-Limb: This cannot possibly have suffered too much analytical thought. The device looks more like it was born out of something else. While it does not always function as maybe intended, it is really lovable. It does not have a reliable precision grip, it is really weak, it tears up its paper thin gloves within minutes.
- TRS prosthetics: Bob Radocy as end-user himself developed by far the greatest useful solutions. But they are not the result of extensive academic efforts, so they cannot be compared to the C-Leg. They are extremely good though and any analysis must start there.
- Toughware PRX: These devices are extremely well made, mechanics wise – but we lack an analytical model that precedes the engineering there as well, comparing this to the C-leg approach.
- Becker Mechanical Hand: Also the Becker hand was clearly built by someone with great practical and pragmatic understanding. No analytical effort of the magnitude of a C-Leg preceded it though.
- Hosmer hooks: they came out of a practical development, no scholarly work appeared to be prepared for these either.
With this, text is cited and summarized from [1].
Introduction
It describes research work that was done between 1995 and 1998 at the Institute of Robotics at the Swiss Federal Institute of Technology in Zurich, Switzerland (ETH Zurich). The research was performed as an EUREKA project (EU-1455) and was partially supported by Otto Bock GmbH Austria, by Kistler AG, and by Swiss KWF (KTI 3150.1).
It was supported also by Professor Dr. E. Stuessi from the Laboratory of Biomechanics of the ETH. Mr. P. Gammer, general manager and Dr. H. Dietl, head of development, of Otto Bock GmbH Austria, provided financial but also technical support. Also, C. Calame of Kistler AG provided support. Furthermore, other people are mentioned in the introduction of this thesis.
Summary
This study first examined the normal and the prosthetic gait as a preliminary study. A dynamic model of both normal and prosthetic gait was developed. The sensitivity of specific parameters and gait determinants was simulated and studied.
From that dynamic model, optimal mechanical impedance (dynamic dampening and stiffness) of the prosthetic knee joint was obtained.
So in other words, this work built its prosthetic knee control of a very extensive effort trying to actually understand and model a targeted function.
The control system contains a number of sub-functions.
- Identification of current state of gait, of walking motion, based on sensor data.
- Gait control, which generates knee flexion and extension dampening levels, according to present and past gait states.
- Hydraulic control, which generates the hydraulic valve angular position, and gait speed identification based on sensor measurements.
- Gait speed identification, which influences both the gait control and the hydraulic control.
It also contains a database of prior and current signals from the prosthesis. Parameters, models and rules are combined to achieve a knowledge base of the system.
Chapter 1 – Introduction
The first part describes principles of normal gait. For that purpose, time constraints in the relative coordination of both legs are placed on a time axis. Time sharing for single and double leg support is considered.
A review of current prosthetic knee joints and of literature regarding knee controllers is performed.
From there, new requirements for the controlled knee joint as developed by the author and his team are laid down:
- sufficiently stable to support body weight
- absorb ground impact at heel contact and allow smooth forward progression with sufficient knee flexion
- body support ona bent knee during mid stance
- instant response to walking speed change, mainly during swing
For that purpose, a first step (pun, sorry) is to model normal and prosthetic gait.
Chapter 2 – Modeling normal human gait
There, multi-body dynamics is used to model human locomotion.
As no general flexible dynamic gait model is available, building on sufficiently optimized design procedures, the author and his team realized they had to provide their own.
They basically use equations to simulate the kinematics, and from these, they then identify the relevant inertia forces and torques as well as generalized active forces and torques to complete the model description with EoM (equations of motion).
To make the complex task, that was implemented in a software tool, a bit easier, a set of simplifying assumptions was put in place:
- all motions in sagittal plane only
- all links are rigid and so all links dimensions are constant throughout gait cycle
- both legs are identical, so also their motion is symmetric with a constant time delay
- each link has a lumped mass located at its center of mass, and a lumbed mass moment of inertia about its center of mass
- joints provide ideal hinge or pin points with pure rotation, and all joint reaction forces act through the joint’s center of rotation
- the stance leg has zero slippage on the ground (no dissipating forces, such as viscous or Coulomb friction exist)
- muscles apply pure torque around each joint, to increase or decrase the relative angle between the two links at that joint
The further chapters focus on exactly that:
- Chapter 3 – Five Link Bipedal Model in sagittal plane
- Chapter 4 – Seven Link Bipedal Model in sagittal plane
- Chapter 5 – Six Link Bipedal Model in sagittal plane
Chapter 6 – Modeling Normal Human Gait
Gait lab data with 17 3D translational markers had been used to fit polynoms and analytical derivatives. The best results were obtained after smoothing the measured data a bit.
Reducing DOF to allow for more robust simpler and thus better optimization was achieved through several link count models.
Chapter 7 – Introduction to prosthetic gait
The goal of this part was to basically investigate passively constructed above knee prostheses for their hip torque.
In essence, the absent eighteen muscles have to be compensated for in their various roles to enact symmetric gait.
Technically, controlled damping, or damping control, was judged to be the most relevant asset in a new prosthetic knee joint. So the active muscle control characteristics of normal human gait had to be translated into a set of damping control parameters for the prosthetic knee.The effect that the dampening then has on the other joints, in terms of torque, was modeled and calculated.
Chapter 8 – The optimum mechanical impedance of a prosthetic knee joint
Chapter 9 – The optimum mechanical impedance of a prosthetic ankle joint
This was both solved mathematically and then simulated. Simulation was evaluated for best trajectories and best hip moment of forces. Two walking speeds were simulated, 1.58 and 2.04 m/s.
While the knee certainly was targeted for sensorimotor control of impedance, the ankle performed worse for constant mechanical parameters than it may have given active controls. So even though constant stiffness springs and rubber inserts are a concession to affordable practical damage control, constantly stiff mechanics seem to underperform in any simulation that attempts to optimize prosthetic leg gait performance to maximize on symmetrical gait.
Despite these insights, parameters were found for optimal design or specification of passive ankle and foot.
Chapter 10 – Prosthetic gait
Simulation proved successful howevermuch increasing degrees of freedom massively increased computation times even on fast computers.
Chapter 11 – Introduction to A/K prosthesis control
Design of this intelligently controlled prosthetic device was split into 4 phases:
- hardware definition, hardware build, software procedures to include the knowledge base, i.e., data base and rule base, execution of the inference
- knowledge base is filled with existing preliminary data and rules by experts who well know the system and required behaviour
- challenge of system with real trials, have expertts modify the knowledge base
- self learning and self tuning of individual parameters is to be introduced and developed
Simulation of biological processes for prosthetic legs need to simulate neural control also of gait. From a good simulated model, possibly, new insights might be gained.
Atrophied muscles in an amputee need lighter designs for prosthetic limbs. Weight and weigjt distribution play a large role in implementing an ideal prosthetic leg.
The leg performance of the prosthesis individually depends on how well the hip torque is transferred to the limb, and how well ground reaction forces, inertial forces and gravitational forces are transmitted to the stump. Prosthetic control is to improve these properties.
Chapter 12 – Description of a prototype A/K controlled prosthesis
The prototype was directed towards a future product considering reliable function, low cost, low weight, low volume, and availability.
Thus the prosthesis consisted of commercially available low cost catalog parts, as a design requirement:
- prosthetic foot
- prosthetic socket
- servo controlled hydraulic damper
- sensing system
- computer
- electronic system
- mechanical construction
The prosthesis will measure all walking, continuously. All possible prosthesis states are observed. The output is the command to the hydraulic damper (impedance generator) in the knee.
Further chapters are entirely technical to implement the aforementioned concepts:
- Chapter 13 – Prosthetic gait control
- Chapter 14 – Finite State Control
- Chapter 15 – Fuzzy Logic Prosthetic Control
- Chapter 16 – Summary
Recommendations for prosthetic arm research
Take a step back. Breathe. What exactly was it you wanted that device to really perform?
- The model of an assisting device needs to be precisely specified.That model needs to clearly specify relevant performance aspects, function and so on.
- It must lean on a real world observation of bimanual task work and its actual requirements.
- If it can be simulated virtually, then instances of realizations that are computerized may be checked for relevant aspects.
- We can also, qualitatively, describe such aspects that would have to go into a succesful assistive gripper model right now, and apply these criteria to evaluating existing solutions.
[Bibtex]
@phdthesis{zlatnik1998intelligently,
title={Intelligently controlled above knee prosthesis},
author={Zlatnik, Daniel},
year={1998},
school={ETH Zuerich, Switzerland}
}