Authors:
(1) Jorge Francisco Garcia-Samartin, Centro de automatic Yar Robotica (UPM-CSIC), University of Politecnica de Madrid-Consejo Superior DE Investigaciones Cientıficas, Jose Josier Abiasal 2, 28006 Madrid, Spain (Spain)[email protected]);
(2) Adrian Rieker, Centro De Automatica y Robotica (UPM-CSIC), Universidad Politecnica De Madrid-Consejo Superior de Investigaciones Cientıficas, Jose Guterres Abiasal 2, 28006 Madrid, Spain;
(3) Antonio Barrientos, Centro De Automatica Y Robotica (Upm-CSIC), Universidad Politecnica de Madrid-Consejo Superior de Investigaciones Cientıficas, Jose Guterres Abiasal 2, 28006 Madrid, Spain.
Links table
Abstract and 1 introduction
2 relevant business
2.1 Air operation
2.2 Aerobic weapons
2.3 Control of soft robots
3 Paul: Design and Manufacturing
3.1 Robot design
3.2 Choose materials
3.3 Manufacturing
3.4 operating bank
4 Gain data and control the open episode
4.1 Device Preparing
4.2 vision capture system
4.3 Data set generation: table -based models
4.4 Open ring control
5 results
5.1 final version of Paul
5.2 Analysis of the work area
5.3 Perform the models based on the table
5.4 Bending experiments
5.5 Weight experiences
6 conclusions
Finance information
A. Experiments and references
4.3 Data set generation: table -based models
Due to the complexity of the robot, models -based methodologies, such as PCC or those that depend on the theory of Cosserat. Although the use of FEM is a way that will not be closed in future work, the large number of parameters to be appointed experimentally each piece (Young Laboratory, the moment .
The system’s product is taken as a position and guidance that are finally reached – at this stage, all the sites of the intermediary parts – as an input, and inflation times for both the mines. Due to the lack of sufficient compression sensors available at the time of the robot construction, it was decided to take the time of inflation as an input variable. Since the work pressure is limited to the pressure reduction valve and it can be assumed that the flow rate in each bladder is fixed, the time is equivalent to the size of the air inserted in each cavity.
All controlled control options share the need for a large amount of experimental data, which leads to the need to develop a experimental design to regulate the collection of these data. Since capturing this information takes place at different stages and data groups should represent robot behavior in an objective way, the re -application of the experiment takes special importance.
The data stored in the data groups was the location of the robot and the set of inflation times that achieve this configuration. The above registration is that only two of the three bladder in the part swelling reduces repetition. As we mentioned earlier, more than two installments lead to repetition, which means that the reverse motor model of the robot can have multiple solutions.
The data collection process includes several successive steps. Initially, a specific number of samples is determined. For each sample, Matlab orders send a random set of nine times inflation, corresponding to each Paul valve, to the operating seat. The times are created less than the maximum TMAX time, and guaranteeing the inflation of the cavity only in the part. After that, robotbbes are amplified based on their transmission times. Next, two vision cameras take two pictures to determine the position and the end of the robot. This entire procedure is repeated for the specified number of repetitions, and upon completion, the data collected in the data set is stored in the data set
Information related to swelling times is stored as a percentage, at a value of 0 zero swelling for this part and 100 corresponding to TMAX, swelling of the maximum number of millimeters specified for this data collection session. This TMAX value, along with values, is stored in the data set, in order to be able to compare different data collections. The reason for this coding is a lack of information, which is prior, at the maximum pressure recognized by Black Black. Although it is correct that inflation times more than 1500 milliliters were determined in a row, which led to holes, the application of the lower times during a frequent number of courses has also been created. On this basis, it was decided not to amplify any valve, either in one or several steps, more than 1000 milliliters.
Along with inflation times for each bladder, the position and direction that was reached by the final end is stored, based on camera readings. In particular, the position of the green mark is stored and directed to Triedron. The latter is expressed in EULER angles, as it is a form of effective storage from the rotating matrix. In addition, the data set also contains identification data from the assembly process that is believed to affect the results, such as the pressure of the air line or the surrounding temperature.
Some aspects of the air system deserve attention. Initially, bladder and contraction enlargement are not similar. Engineering restrictions in the air components lead to a low contraction rate compared to inflation. Thus, when Paul receives shrinking time, he doubles it with an experimental derivative factor, about 1.45 for work pressure 1.2 bar. This double compensates for the contradiction between inflation and contraction times for an individual group of bladder, ensuring that the contraction time is in line with the time required to reach the same inflation point.
Likewise, although it is physically possible to amplify many valves at the same time, it has been shown that the distribution of parallel flow means that the active fillings of each valve are not the same as as if they were individually inflated. To prevent this phenomenon, it was decided to amplify each bladder individually during the process of obtaining data and when she was asked to reach certain sites.
Finally, there are the slowdown in silicone that causes the position reached by amplification for a period of time to be different from the position that was reached by amplifying the time for a period of T1 and then for a period of time T2 = T – T1. The strategy used to address this problem was to capture the data group that brings Paul to his zero position between each sample. However, when controlling the robot in the open loop, this is not possible, or at least, unwanted, as one may want to follow the paths or travel through a series of points. Therefore, the transition from the X1 position to X2 requires an additional factor of 1.2, also experimental, to calculate the effective effects.
4.4 Open ring control
Once you create a data set, it can be used to model Paul’s behavior to control the open loop. It is expected, as a future line, to train a nervous network on direct movements and other reverse mobility. However, given the amount of large data that may be required (in [62] 24389 samples are used for a three -part robot like this one), how to search for the table for this work is used.
The direct movement method – which allows to obtain a place and direct the end end of the robot from the inflation times in the capacity – from the search, in the set of data created in the previous step, and the values of the three inflation time in a shorter distance from the time of inflation given as a reference. Obviously, if the set of inflation times required in the table, the value associated with these times will be returned as a result of the direct motor model. Other than this, the average position and direction associated with the nearest three -time inflation, weighed with the distance (the regional standard) that exists between each of them and the values of reference inflation times, such as location and value of the robot direction.
With them, it is possible to calculate the position that is returned by the direct motor model using the expression:
