This page describes the data set generated by running the experiment. One such data set is provided in the 'data_set' folder of the downloadable archive for the experimental paper. This data set is the basis for results presented in the description of the experimental paper.
The trial-level results represent the performance of the participants in terms om times and errors. These are stored in results files whose names start with 'results ...'. The format for the results file names are as follows:
results_su_[subjectID]_se_[sessionID]
See [AdaptiveCollisionHand-Code-Startup] for [subjectID] and [sessionID] descriptions. The header denoting the content of each column (not included in the files for the data set provided) is as follows:
Execution,Trial,Attempt,Error,Time,Time Stamp
where 'Execution' refers to a counter for all successful trials, from 1 to 25 for each participant for each session. 'Trial' is an indicator of the trial (target location) used, from 1 to 5. 'Attempt' is a counter of the number of attempts at a given trial for a given execution. That is, until it is successful and the 'error' for that trial and that execution is '0'.
'Time' indicates the time from the participant pressed the start button until the task was completed, in seconds, while 'Time Stamp' is the current computer UNIX time, also in seconds. The latter can be used to correlate the development of the neural network weights (also time stamped) with the performance of the participant.
The trajectories contain all time-varying data within each attempt, and are stored in files that begin with 'trajectory ...'. There are two trajectory files for each attempt.
For the shared control output, the format for the trajectory file names are as follows:
trajectory_su_[subjectID]_se_[sessionID]_e_[execution]_t_[trial]_a_[attempt]_uci
where [execution], [trial] and [attempt] correspond with the 'Execution', 'Trial' and 'Attempt' lables used in the trial-level results. The header denoting the content of each column (not included in the files) is as follows:
Time,x,y,z,roll,pitch,yaw
Where the each time-varying value is the velocity (not position) in the respective Cartesian components that is commanded by the shared control. The rotations are included in the trajectory files, but are fixed during the experiment.
For the remaining output, the format for the trajectory file names are as follows:
trajectory_su_[subjectID]_se_[sessionID]_e_[execution]_t_[trial]_a_[attempt]
where [execution], [trial] and [attempt] correspond with the 'Execution', 'Trial' and 'Attempt' lables used in the trial-level results. The header denoting the content of each row (not included in the files) is as follows:
Time,x,y,z,roll,pitch,yaw,dx,dy,dz,droll,dpitch,dyaw,
nx,ny,nz,nroll,npitch,nyaw,ux,uy,uz,uroll,upitch,uyaw
Where 'x','y','z','roll','pitch' and 'yaw' are the current Cartesian components of the position. The 'dx','dy','dz','droll','dpitch' and 'dyaw' labels are the current Cartesian velocities sent to the shared control. The 'nx','ny','nz','nroll','npitch' and 'nyaw' labels are the components of the noise received. The 'ux','uy','uz','uroll','upitch' and 'uyaw' labels are the Cartesian velocities received from the user. The rotations ('roll', 'pitch' and 'yaw') are included in the trajectory files, but are fixed or not used in the experiment.
The neural network weights at a given time is stored in a file with name format:
w_su_[subjectID]_ti_[timeStamp]
where '[timeStamp]' is the computer UNIX time, in seconds. The list of timestamps used for the neural network weight files for each subject can be found in the files with name format:
list_w_su_[subjectID]
The neural network weights are stored as a 20 by 20 matrix, with numbered rows. Each row corresponds to a neuron representing the virtual sensors, while columns the neurons representing the proximity sensors.
Three different analysis examples are provided. All are Matlab scripts, tested on Matlab 7.11.0 (R2010b). No external toolboxes are required. Please run from the 'analysis_examples' folder, or change the relative paths of the respective files in the code.
The Matlab script 'create_result_statistics.m' loads the respective trial-level results files and plots the average and the standard error of the Mean Time (MT) and the errors (over participants). See the experimental paper description.
The Matlab script 'run_load_trajectories.m' loads a given set of trajectories into a cell array by calling the function in 'load_trajectories.m'.
The Matlab script 'create_weight_plot.m' plots the development of the neural network weights over time.
Wiki: AdaptiveCollisionHand-Code-Startup
Wiki: AdaptiveCollisionHand
Wiki: ExperimentalPapers
Wiki: Home
