Making sense of experimental data

[sandipan1]

I am not to understand how to use the experimental data given in docs .In the hamner2013muscle dataset there are data given for 20 subjects .Each folder containing

I am not sure about where are muscle activation from the experiments Is it the CMC results ? if so there are negetive values , positive values (>1) . How do I use that for simulating muscle activation (for the competition)?

[kidzik]

Raw EMG recordings are here https://simtk.org/frs/?group_id=603. However that there is no EMG for all muscles and these are noisy signals in general. CMC is an approximation of muscles activity given the model and it's probably more useful than raw signals in this case. Note that in the output you have results from many stages of optimization, so the early files have values which don't match conditions. To my understanding, negative values are also a consequence of the fact that the optimization algorithm did not find a result matching all conditions (@carmichaelong is it correct?). For RL it might be enough to just assume that negative values are zeros and start to search around these 'incorrect' trajectories.

[carmichaelong]

Some notes about the CMC results folders:

• The file ending with controls.sto has the data that is roughly equivalent to the inputs in this competition (the input is called excitation). They are found in the 2nd through 93rd columns (glut_med_1 to extobl_l) for each of the 92 muscles in the model (many more than we're using). All of these columns will have values between 0 and 1. The remaining columns are extra forces/torques used to help solve the CMC problem since the data captured is noisy and may not match real life perfectly (these columns can have any value).
• Another file to note is the file that ends in states.sto. These have the corresponding state information that goes along with the controls.sto file. Although some variables here end with .activation, these do not correspond to the input of the challenge. These represent the actual muscle activation, which is (roughly) a delayed signal after the excitation.

[sandipan1]

@kidzik @carmichaelong I listed the muscles in control.sto and tried to match it with ProstheticsEnv observation muscles. I couldn't find muscle"abd_r" ,"abd_l" "hamstrings_r" , "hamstrings_l" in controls.sto There are "add_long_r" "add_brev_r" "add_mag1_r" "add_mag2_r" "add_mag3_r" along with their left leg counterparts in controls.sto .However there is only"add_r"in the muscle set. How do I choose? Similarly "glut_max1_r", "glut_max2_r", "glut_max3_r" in control.sto but only "glut_max_r" in muscle set "vas_med_r", "vas_int_r", "vas_lat_r" in controls.sto but only "vasti_r"in muscle set "med_gas_l", "lat_gas_l" in controls.sto but only "gastroc_l"in muscle_set

How do I map these muscles?

[kidzik]

The CMC output wouldn't produce the movement anyway, so you can treat it as some approximation and use it as a reasonable starting point for the reinforcement learning (i.e. trying to first build a network in a supervised manner). You can also try to rerun CMC with the model from the challenge.

[sandipan1]

Okay.Thanks

[ThGravo]

@sandipan1 Did you manage to rerun the CMC?

[sandipan1]

Nope .I tried to change the setup files and run but the execution got stuck while running CMC tool.