Implementation of Static Optimization

[Erik Dijkstra]

Hi all,

I would like to perform an optimization with the OpenSim model using the API, preferably through MATLAB. However, I could not find out how I get all the necessary data from the model.

My thought was to start from the Static Optimization from the Analysis Tool but I cannot find its implementation file. I can find the header files but I would like to get a clear view of how to get the system parameters necessary for:
- contstraintFunc
- objectiveFunc (in Analyse Tool, the computation of muscle activations squared)
- gradientFunc
- constraintJacobian

These functions should be described somewhere for the Static Optimization Tool as well but I can't find them in any file.

I realize that this might, in its fundamentals, be more of a SimTK problem but since it is used in OpenSim and my application would be for use with/in OpenSim I decided to post it here.

Basically:
Where I can find the implementation files for the Static Optimization(Not the two header files) such that I can create my own Optimization (Tool) where I have additional/different optimization variables and constraints. Hints on how to access the system properties necessary for Optimization are also welcome.

Best regards,
Erik

[Daniel Krueger]

Hey Erik,
perhaps the following c++ code may help you:

Code: Select all

               State s=this->model.initSystem();
		Storage statesStore;
		statesStore.setDescription("initial pose sequence");
		Array<std::string> stateNames;
		stateNames=this->model.getStateVariableNames();
		stateNames.insert(0, "time");
		statesStore.setColumnLabels(stateNames);
		Array<double> stateVals;
		this->model.getStateValues(s,stateVals);
		//Fill state storage with a sequence of copies
		for(int i=0;i<this->hz;i++){
			double time=t0+i*this->dt;
			statesStore.append(time,stateVals.size(),stateVals.get());
		}
		statesStore.setInDegrees(false);
		//Static optimization 
		StaticOptimization so(&this->model);
		so.setStatesStore(statesStore);
		double ti, tf;
		statesStore.getTime(0,ti);
		statesStore.getTime(this->hz,tf);
		so.setStartTime(ti);
		so.setEndTime(tf);
		//run the analysis
		for(int i=0;i<this->hz;i++){	
			//copy states
			statesStore.getData(i,s.getNY(),&s.updY()[0]);
			double t=0.0;
			statesStore.getTime(i,t);
			s.setTime(t);
			this->model.assemble(s);
			this->model.getMultibodySystem().realize(s, SimTK::Stage::Velocity);
			if(i==0){
				so.begin(s);
			}else if(i==this->hz){
				so.end(s);
			}else{
				so.step(s,i);
			}
		}

		//write the results into control vector
		for(int i=0;i<this->hz;i++){
			this->u[i]=Vector(this->szU);
			so.getActivationStorage()->getData(i,this->u[i].size(),&this->u[i][0]);
		}

Best regards,
Daniel

[Erik Dijkstra]

Daniel thank you for sharing your code. I think it will help me later on to get data from/feed data to the optimizer at intermediate levels and not just at the end.

In the mean time I think I found the code which can help me to build a modified optimization. I downloaded the source code for OpenSim, here more details are shown in the cpp files; StaticOptimizationTarget.cpp and StaticOptimization.cpp (compared to the header files accessible in standard OpenSim installation).

Best regards,
Erik