• Make a new file in Data_json/[new_folder]/[new_trial.json]. These json files sets up the run configurations and specific files to load. So it applies "runParamGlobal" to all the settings, then gets overwritten with "runTemplates", then gets overwritten with "Files". You can see the final output in inverseIOC/ioc_main.m line 58, in trialInfo. This sets the parameters for what to load.

• In InverseIOC/IOCRun.m, you see that the line 5 loads model. This getmodel file is in logic/getmodel.m. You need to add an entry that matches trialInfo.baseModel (the experiment type) and trialInfo.model (any sub models you may have) to the switch function.

• Then make a copy of https://github.com/adaptivesystemslab/ioc/blob/master/Logic/ModelRL_Template.m and update this appropriately. It's mainly making sure the joint angle array are loaded properly, as well as any kinematic/dynamic models.

• To add new cost functions, you need to name it in common\featuresEnum. This name is what trialInfo is looking for. Then just add a corresponding entry in featuresCalc. If you look in featuresCalc.initCf, you will see that each enum has a set of basis fucntions (bf), and the actual cost function (cf) itself. The basis function is the fundamental feature required. So how to go from the model kinematics/dynamics to the cost function. It's listed here so it gets calculated at each iteration, to save computational effort. You can either just use a inline function (ie featuresEnums.cartVeloSumSqu), or write your own function if the cf calculation has a few steps (ie featuresEnums.cartCurvatureSumSqu)

• The gradient should be calculated automatically, but you can see the results in inverseIOC\IOCRun.m line 76, where the cost function values (called features here) and the model dynamics are gradients are being calculated

Once this is done, you should be able to run the function, and graph the outputs using `analysis\iocAnalysis00_plot_demo'