mathias wellner

Today the second Loko­mat Sym­po­sium fea­tured the main talks regar­ding the Loko­mat actua­ted gait ortho­sis. Actually the main sci­en­tists were all joi­ned to pre­sent their latest rese­arch. To give you a short impres­sion I will list the talks.

I did not have a talk but was invol­ved in some lab demos of the vir­tual environ­ment exten­sion of the Loko­mat and the addi­tio­nal degrees of free­dom (Gait3D). The audi­ence was really impres­sed and espe­cially for child­ren the vir­tual rea­lity enhan­ce­ment trig­ge­red wild interest.

I am in! After sen­ding an abstract to the con­fe­rence Medi­cine Meets Vir­tual Rea­lity about the Loko­mat and vir­tual rea­lity (obst­a­cle cros­sing), I now recei­ved the mes­sage, that it was accep­ted for pos­ter pre­sen­ta­tion. And so my name appears once more in the www, on the Abstract Gui­de­lines page. It’s there, at the very, very end of this long list.

Ges­tern (21. Okto­ber) fand im Sem­pert­schen Haupt­ge­bäude der ETH das ZNZ–Sym­po­sium statt. Das Zen­trum für Neu­ro­wis­sen­schaf­ten Zürich ver­sucht, die ver­schie­de­nen For­schungs­grup­pen an der ETH Zürich und an der Uni­ver­si­tät Zürich im Bereich der Neu­ro­wis­sen­schaf­ten zu ver­net­zen. Höhe­punkt ist sicher­lich das Sym­po­sium mit Vor­trä­gen und Poster-​​Sessions.

Ich fand’s ziem­lich lang­wei­lig, weil ein Groß­teil der unter dem ZNZ-​​Dach zusam­men­ge­fass­ten For­schungs­grup­pen auf der mole­ku­la­ren und bio­lo­gi­schen Ebene arbei­tet. Inge­nieure sind ganz klar in der Min­der­heit. Zur Unter­strei­chung die Titel eini­ger Vorträge:

• The shaky life of memory: Com­pe­ti­ti­ons, desta­bi­liza­ti­ons, and the bio­logy of the long-​​term trace
• Rho Gtpase func­tion in myelination
• Nuclear signal­ling by the amy­loid pre­cur­sor protein
• Effect of pre­na­tal infec­tion on GABAer­gic neu­rons in a novel ani­mal model of Schizophrenia
• Single-​​cell and net­work activity in the neo­cor­tex in vivo
• Explo­ring the brain, from beta probe stu­dies in rats to PET expe­ri­ments in humans
• Are adult and plas­tic songs in the zebra finch gene­ra­ted by dif­fe­rent neu­ral mechanisms?
• Keynote Lec­ture: GABA and the regu­la­tion of emo­tion and cognition

So sah’s also aus — unver­ständ­li­che Abkür­zun­gen, schöne bunte Gehirn­gra­fi­ken, wel­che die Prä­senz irgend­ei­nes Mar­kers zeig­ten und mit­ten­drin ich, auf der Suche nach Men­schen, die ich ver­stehe. Das Inter­esse zum gegen­sei­ti­gen Beschnup­pern war dann auch eher gering, ich beglückte einen Neu­ro­in­for­ma­ti­ker damit, dass ich mich für sein Pos­ter inter­es­sierte. Und das, obwohl sein Thema (Selbst­or­ga­ni­sa­tion von zel­lu­lä­ren Ein­hei­ten zur Nach­bil­dung der neu­ro­na­len Ver­net­zung) gar nicht so unspan­nend war. Am Ende gab’s dann einen Apero, mit sehr auf­merk­sa­men Catering-​​Damen, die stets mit Wein­fla­schen herum lie­fen und alle Glä­ser auffüllten.

One of the first steps I had to take was under­stan­ding human gait. The arti­cle (biblio­gra­phic infor­ma­tion) descri­bes very ela­bo­rate model­ling tech­ni­ques in order to make a mecha­ni­cal gait ortho­sis walk. I would like to empha­size that th Loko­mat is not inten­ded to make pati­ents walk around, so it’s a dif­fe­rence to the device men­tio­ned in the arti­cle. But the gene­ral mecha­nisms are com­pa­ra­ble, espe­cially using tor­ques to make the pati­ent take a step.

Abstract — The pur­pose of this study was to examine a hybrid ortho­sis sys­tem (HOS) for wal­king after spinal-​​cord injury (SCI) that coor­di­na­tes the mecha­ni­cal locking and unlo­cking of knee and ankle joints of a recipro­ca­ting gait ortho­sis (RGO), while pro­pul­sive forces are injec­ted and unlo­cked joints con­trol­led with func­tio­nal neu­ro­mus­cu­lar sti­mu­la­tion (FNS). The likely effec­tiv­en­ess of the HOS in terms of for­ward pro­gres­sion, sta­bi­lity, and pos­ture of para­ple­gic gait was deter­mined in this simu­la­tion study. A three-​​dimensional com­pu­ter model of a HOS com­bi­ning FNS with an RGO incor­po­ra­ting feed­back con­trol of mus­cle activa­tion and joint locking was deve­l­o­ped. An anthro­po­mor­phic human model inclu­ded pas­sive joint moments and a foot-​​ground con­tact model adap­ted from other stu­dies. A model of the RGO recipro­cally cou­pled the hips and locked and unlo­cked the knee and ankle joints during stance and swing respec­tively. The actions of mus­cles under FNS activa­tion were mode­led via closed-​​loop con­trol of joint tor­que inputs. A wal­king aid that mimi­cked canes and volun­tary upper extre­mity actions main­tai­ned late­ral sta­bi­lity by pro­vi­ding the necessary shoul­der forces and moments. The simu­la­ted HOS achie­ved gait speeds of 0.51/spl plusmn/0.03 m/​s, stride lengths of 0.85/spl plusmn/0.04 m, and caden­ces of 72/​spl plusmn/​4 steps/​min, excee­ding the repor­ted per­for­mance of other assis­tive gait sys­tems. Alt­hough mini­mal for­ward trunk tilt was found to be necessary during spe­ci­fic pha­ses of gait, pos­ture, and sta­bi­lity were signi­fi­cantly impro­ved over FNS-​​only systems.

The arti­cle (see biblio­gra­phic infor­ma­tion) fits very much in what I am plan­ning to do with the Loko­mat. It is very well done in my opi­nion and a good base for my work. A robot with three degrees of free­dom is used to inves­ti­gate the pro­po­sed iden­ti­fi­ca­tion and con­trol methods.

Abstract—A com­bi­na­tion of model-​​based and ite­ra­tive learning
con­trol (ILC) is pro­po­sed as a method to achieve high-​​quality motion
con­trol of direct-​​drive robots in repe­ti­tive motion tasks.We include
both model-​​based and learning com­ponents in the total con­trol
law, as their indi­vi­dual pro­per­ties influ­ence the per­for­mance
of motion con­trol. The model-​​based part of the con­trol­ler com­pen­sa­tes
much of the non­linear and cou­pled robot dyna­mics. A new
pro­ce­dure for esti­ma­ting the para­me­ters of the rigid body model,
imple­men­ted in this part of the con­trol­ler, is used. This pro­ce­dure
is based on a batch-​​adaptive con­trol algo­rithm that esti­ma­tes the
model para­me­ters online. Infor­ma­tion about the dyna­mics not covered
by the rigid body model, due to fle­xi­bi­li­ties, is acqui­red expe­ri­men­tally,
by iden­ti­fi­ca­tion. The models of the fle­xi­bi­li­ties are used
in the design of the ite­ra­tive learning con­trol­lers for the indi­vi­dual
joints. Use of the models faci­li­ta­tes quan­ti­ta­tive pre­dic­tion of per­for­mance
impro­ve­ment via ILC. The effec­tiv­en­ess of the com­bi­na­tion
of the model-​​based and the ite­ra­tive learning con­trol­lers is
demons­tra­ted in expe­ri­ments on a spa­tial serial direct-​​drive robot
with revo­lute joints.