Eurobench template 1. Introduction The addition of new protocol to the Eurobench Benchmarking Software requires providing a set of information for filling the database. An excel sheet is provided to gather that information. All tables are followed by an explanation of each of the fields that deserves it. 2. Scenario sheet Table 1. Scenario Template sheet Item Template designer description short name Description Image Assessed System Abilities short name: single sentence scenario description. Likely to be used for filtering purposes in the front-end to help the user find the appropriate protocol. description: extended description of the scenario. Considering structured text (markdown or adoc), possibly accepting latex equations. image: single image that can be provided to visually illustrate the scenario Logo are accepted. 3. Protocol sheet The table is described per block just for easing the description of it. Table 2. Protocol Template sheet - Generalities Item Template designer description name Definition Image Keywords Suitable Bipedal systems Associated PI Algo Name: single sentence protocol description Definition: General description of the protocol (like a sentence). Considering structured text (markdown or adoc), possibly accepting latex equations. Image: illustrative image keywords: filtering terms that can be associated to that protocol. Can be related to the action asked to the subject or testbed characteristics… Suitable bipedal systems: definition of the subjects for which the protocol is designed. Can only contain values like [humanoid, prosthesis, exoskeleton, human], where human stands for a human subject not using any robotic device. Associated PI algo: List of software routines that are to be used to score any experiment following this protocol. Several can be provided (considering that a PI algo can compute several PI scores). Should be related to the short name provided within the [PI Algo sheet]. Table 3. Protocol Template sheet - Controlled variables name Definition type range unit … … Controlled variables are conditional settings that should be filled by an experimenter, to describe their experimentation conditions. These variables are not supposed to change during an experiment trial execution. Parameters can be related to either: a configuration parameter of the testbed (for a walking on slope protocol, the controlled variables could be the slope angle and the slope length). a configuration of the robotic system (like setting different levels of assistance of an exoskeleton, or testing different set of tuning parameters of an humanoid) a behavior indication provided to the subject (like do the action with eyes closed) The controlled variables, together with the protocol spec, should enable anybody to reproduce a given execution with exactly the same settings. Name: single word. If needed under_scored like name can be used. Definition: text description of the protocol. type: string, integer, float… range: [minimum, maximum] value the variable can have unit: if any, variable unit (preferably using the International System of Units convention). Table 4. Protocol Template sheet - Testbed Item Type (Sensor, actuator or combined) Technical file sheet … … Each item that can be bought and installed without customization have to be listed (e.g. motion capture system, IMU, force platform, …). If a specific instrumented equipment has to be constructed, it has to be mentioned as a standalone item, even if it includes several sensors and or actuators. The technical file sheet should provide the BOM (Bill of Materials), and the mounting instructions. Example of entry could be instrumented chair, or sensorized wall. Table 5. Protocol Template sheet - Protocol Description General description Step Description 1 … 2 … As a recipe, the Protocol description should provide all the successive operations should conduct the experimenter to conduct an experimentation, from the system preparation, subject preparation to the execution and data collection. The field General description should provide a short (e.g. one sentence) overview of the action taking place during the experimentation. This table may not contain enough information to detail all the steps (like how and where to place markers on human bodies). We assume more exhaustive description will be provided with the documentation associated to the protocol, aside that excel sheet. 4. PI Sheet First we start reminding two concepts (this should be brought to a dedicated page): A Performance Indicator (PI) can be associated to an outcome parameter obtained used to score an experimentation. It can be related to a physical dimension we want to characterize, or to a subjective (based on human scoring) evaluation of a system. A PI can be computed from different input information (motion capture, IMU …), and thus be related to different types of code (or PI Algo).. A Performance Indicator Algorithm (PI Algo, also referred to as benchmarking algorithm, or software routine, or processing routines) corresponds to a specific code implementation that is able to compute one or several PIs. The PI Algo is naturally associated to specific input information that is needed for making the computation (or run the algorithm). Table 6. Performance Indicator sheet Item Template designer description name Description Unit output_type intra_run aggregation inter_run aggregation name: descriptive short name (using underscore). This name should be use the PI Algo for storing the corresponding the outcome parameter when executed. Description: Extended description of the PI (should explain what it is about) Unit: If any output_type: Related to the output file containing the PI score. Selection in default type defined: [scalar, vector, matrix]. intra_run aggregation: How to compress the PI obtained in one run into a more compact representation. Selection in default type defined. inter_run aggregation: How to compress several PIs obtained across different runs into one representative PI of the entire experiment. Selection in default type defined. Some item deserves more detailed explanation: 4.1. output type field The output type field refers to the type of output file used to describe the Performance Indicators. The type provided must refer to one of the output types described in the PI specification file. For now, format accepted are: scalar, vector, matrix, vector_of_vector, vector_of_mtrix and string. 4.2. intra-run aggregation In the previous example, we see that the output is represented by a possibly long vector. The Intra-run aggregation information provides the system with an operator tool for “compressing” that vector into an (hopefully) more digest information. If the mean operator is selected the PI Manager will adjust the previous file to contain the following extended information: type: 'vector' value: [0.96867, 1.01667, 0.98843, 0.95168, 0.87936, 0.94576, 0.87802, 0.87571, 0.81802, 0.82336] aggregations: [[0, 'mean', 0.9145679999999998]] The 0 indicates the row/column to which the aggregation has been applied. Several rows/columns could be specified in the case of an ouput of the type matrix: type: 'matrix' value: [[0.96867, 1.01667, 0.98843, 0.95168, 0.87936, 0.94576, 0.87802, 0.87571, 0.81802, 0.82336],[0.96867, 1.01667, 0.98843, 0.95168, 0.87936, 0.94576, 0.87802, 0.87571, 0.81802, 0.82336], [0.96867, 1.01667, 0.98843, 0.95168, 0.87936, 0.94576, 0.87802, 0.87571, 0.81802, 0.82336]] aggregations: [[0, 'mean', 0.9145679999999998], [2, 'mean', 0.9145679999999998]] That compressed information can now be presented to the user on the Front-End. The currently available operators are related to panda operators, i.e : [mean, median, mode, std, min, max, var, sum, abs, none]. Note that we can define various operators for one PI (like mean and standard deviation). The intra-run aggregation is thus using the following pattern: [ [number, operator] ]. Obviously it does not need to be filled for scalar output. 4.3. inter-run aggregation A similar operation is needed for combining the scores obtained for each run, to generate a PI for the whole experiment. Similar operators are proposed, but they will be computed on all the values obtained for all runs. For instance, if the experiment contains 3 runs, characterized with the PI step_length_left, we would have after executing the PI Algo 3-run PI files: Subject_10_trial_01_pi_step_length_left.yaml, Subject_10_trial_02_pi_step_length_left.yaml, Subject_10_trial_03_pi_step_length_left.yaml. The Inter-run aggregation will enable to generate a PI score for the whole experiment associated to the left step length: aggregations: [[0, 'mean', 1.0224573076923078]] Assuming that the operator mean was indicated in the Inter-run aggregation field. That score could be presented to the user on the front-end, to summarize the score obtained by the experimentation on that concrete PI. The inter-run aggregation is thus using the following pattern: [ [number, operator] ]. It has to be filled for all output format, even scalar (since we still have to aggregate the scalar value of each run). 5. PI Algo Sheet Table 7. Performance Indicator sheet Item Template designer description name Description Url paper Url code Associated PI Input files Input command name: descriptive name description: description of the algorithm used for making the computation url paper: link to a paper describing the algorithm used or closely related (if available) url_code: link to repository code. Could be public or private (additional field?). associated PI: Selection of the different PIs that the algorithm will compute input files: List of input files expected (according to the Eurobench Data Format) input command: Name directly connected to the input files + dependent variables info 5.1. Input command At this moment it contains the entry point and the parameters the entry point is expecting. The format for the string is the following one. entry_point;param_1;param_2;param_N Example: pi_csic: ./run_pi;jointAngles.csv;anthropometry.yaml: entry point = ./run_pi param_1 = jointAngles.csv (it will look for a csv file containing jointAngles in its name) param_2 = anthropometry.yaml (it will look for a YAML which contains anthropometry in the filename) The system will use that pattern to find appropriate files within the experimental data files made available by the experimenter. Note that the PI algorithm is expected to generate as many files as the number of PI associated to it. Furthermore, the name of each result file should be the name of the PI it represents (as defined in the field name of the PI table). With an example: pi_csic algorithm scores 6 PIs for a single run: step_lengh_right, step_length_left, stride_time_left, stride_time_right, step_time_left, step_time_right. So that after executing the pi_csic algorithm on one run, we expect to get 6 files generated: step_lengh_right.yaml, step_length_left.yaml, stride_time_left.yaml, stride_time_right.yaml, step_time_left.yaml, step_time_right.yaml. All files should have a YAML structure, following the structure associated to each PI (as defined in the related field output_type of the PI table).