Newsfilter article unlocker

Hello everyone, welcome and thanks for joining today's webinar. We are really excited to talk about our new Xsense HIP algorithm and motion reference unit, the new standard for marine applications. I am Martina Ravizza, Product Manager at Xsense, and today with me there is Ayush Sharma, Research Engineer, the brain of the HIV algorithm, who will talk about it in technical detail. I will start with an overview about Xsens, and then I will introduce the Xsenses MRUs, also explaining how they can be used in what kind of applications. And then I will leave the microphone to Ayush, who will talk about him. At the end of the presentation, we will host our Q&A session. Some logistics. The webinar will be recorded and the recording will be available online and you will also receive it via email. you can type your questions please use the test the question tab at the bottom and at the end of the presentation we will answer all the questions so exsense is specialized in design manufacturing and sales advanced advanced motion tracking solution we provide accurate orientation and position for various applications and markets was founded in 2000 and we have a global presence with five regional offices more than 140 regional sales and distributors many defining patterns in 3d motion tracking and we are also iso 9001 certified since 2015. Here you can see our industrial portfolio. We offer different product series in different form factors with different interfaces and different functionalities. From IMU to provide calibrated inertial sensor data, but also VRU, AHRS for orientation, roll, pitch and heading, and also inertial navigation system to provide position also with RTK correction, so centimeter level position. On the left, you can see our smallest form factor, the smallest IMU available in the market, including the sensor fusion algorithm embedded in it. It's available in all the functionality from IMU up to RTK INS. And it's really small, meant for deep integration, where the size is a big and important factor for you. Moving to the right, we see our OEM solution, the 600 series OEM and the XS-AVIOR, our latest release. IP 51 housing, plastic housing for the 600 Series and aluminum housing for AVIOR. where the size is important of course the compact low power consumption low weight and available with a more flexible interface uart rs232 depending on the series also can and so on and then we have the rugged version ip68 housing more robust and also in the case of accent series and the MTI G710 military standard certified, so proven to be resistant against high vibration, dust, water and meant for harsh conditions. On the right, the Xsens Vision Navigator is shown, a solution that includes a dual RTK and also a camera to provide position also in in genesis denied environment our products are industrial grade and either free as you see can see on the bottom they are suitable for many different applications to mention some marine applications surface and subsea use cases off-road mobile robotics agricultural robots and vehicle mining trucks, surveying and mapping, also indoor mobile robotics, drones, and any vehicle or system that needs to navigate autonomously. But what's the core of our technology? Here you can see a simplified architecture of our sensors. Each product includes an IMU, so a three-axis gyroscope accelerometer and magnetometer, and depending on the sensor can be internal, inside the housing you can find some aiding sensor or if it's the SMD or OEM form factor is an external aiding, so you can't find the GNSS module, RTK module, dimerometer and a camera. The high frequency crystal MCU takes care of the synchronization and timing of all the sensor inputs and also apply the calibration. Calibration is unique for each sensor and also consider the nonlinear temperature dependency. The MCU also runs the XSENS high rate strap down integration. We are talking about frequency from 1 kHz for the 1-series up to 10 kHz for the X10-serie series, and it is important to be able to measure any kind of dynamics, also high frequency dynamic dynamic movement. And also applied coding and schooling compensation, and these adjust for any acceleration and gyroscope errors. The MCU then runs the advanced XSEN sensor fusion engine that combines all the synchronized, calibrated, and downsampled data and is able to output highly accurate 3D calibration inertial data, 3D orientation, roll, pitch, and yaw, depending on the product, heave, position, velocity, uncertainty output. Is it possible to configure the sensor, selecting, for instance, the output you would like to receive through the MT-Manager software or the SDK? The system can communicate through a wide range of interfaces, depending on the product, from a low-level interface, including SPI, UART, or CAN, RS-232, and so on. It's also possible to synchronize the sensor with other sensors that might be in your system thanks to an external trigger. But what makes our product special? As we have seen in the previous slides, we provide a wide range of product series with different functionalities from IMU up to visual inertial navigation system. and also the products are really flexible, different form factor and interfaces that makes suitable for any needs. Each unit before shipment is factory calibrated, so it must comply with numerous quality indicators, including noise, bias, temperature, and also the magnetometer is calibrated to resist against magnetic interferences. We leverage more than 25 years experience in sensor fusion, so our outputs are really accurate, they are reliable and robust also in the field. And what our customers value a lot is the ease of use of our products. They are plug and play, and they come with a comprehensive free software suite, including an empty manager with a user-friendly interface, a software development kit, including multiple examples in different programming languages, free technical support from our expert field application engineer, and also a strong partner network, including NVIDIA, Jetson, ROS, and many more. In the end, our products have a long life cycle, and we also take care of a keen end-of-life management. But what is new? With the release of the latest HEAD algorithm, we added to our portfolio XS-AVIOR and XS EVIOR's Motion Reference Unit, industrial-grade inertial sensors that are designed to deliver orientation in here in a compact form factor suitable for marine application. More in detail, XS Savior has an OEM form factor with an IP51 aluminum housing, is meant for all the applications where the size, the weight, and the power consumption is important. It is highly accurate and thanks to the advanced analog signal processing pipeline delivers robust data also with high vibration and hard shock. It can offer a wide range of interfaces as you can see here SPI, R2C, RS-232, CAN and so on. XM series is the rugged version so we're talking about IP68 housing and also military standards so it's really meant for challenging environment resistant to shock vibration and also of course process any of our products magnetic in here. Key applications for our MRU are any marine application that requires accurate attitude, roll pitch, heading and heave, including offshore drilling and crane operation, docking and dynamic positioning, buoys, marine robotics, RVs, AUVs, vessel stabilization, surveying and mapping and many more. What is important? We know that the magnetic interferences is a critical aspect to consider in marine application. That's why we initially calibrate each unit internally before shipment. But of course the magnetic environment can varies in the use case scenario. So we provide different tool to recalibrate the magnetometer. We can use, you can use the magnetic film mapper that requires a PC connection and it's needed to slowly rotate the vehicle where the sensor is mounted to map the environment, collect all the magnetic environment data and recalibrate the magneto meter. If it's not possible to have the PC close to the sensor, there are other options, the representative motion and the in-ramp compass calibration. In this case, for instance, for the representative motion, it's an embedded MFM, magnetic field mapping, that runs directly on the sensor. And this is suitable for all the application where it is possible to reproduce the expected dynamics. While the in-run compass calibration runs in background and continuously calibrate the sensor during the operation. Of course, the best approach for the best result is suggested to use the magnetic field mapping. But of course, if there are some constraints, also the other tools are suitable. I would like now to show you our latest video, where you can see how we tested our new MRUs and HIP algorithm, especially in Frieland in the Netherlands, in the North Sea. Enjoy the vision. We are in Fleeland today in the Netherlands and we are testing our new HEAV algorithm for the XSENZ Series and XSENS Aviar MRU. HEAVE is the vertical displacement of a ship or vessel, any system on water that needs to compensate for this up and down motion. I have with me the Xsens Series device which will be supporting HEAVES estimation. For reference, I have the Xsense 680G with RTK support, which we'll be using for ground fruits data collection, and then verify our EV algorithm accuracy. Shortly from now, we will be heading to the north of Flea Land, where we expect waves up to 150 centimeters, so that's sea state four. We tested different kinds of scenarios. We went with the waves, we went against the waves and also for a very long time we were standing still in the ocean. And in all the test cases that we encountered, in all of the wave periods, in the wave heights, we measured in all cases less than 5% of error. In the regions of really long wave periods ranging between 38 to 40 seconds, we measured up to 6% of accuracy. This range is already much, much broader than what is available in the market. So we are really happy with the results and confident in releasing a great product for our marine customers. All right. I hope you enjoyed the vision. We had fun. It was really nice to see our MRU in action. And yeah, as mentioned already in the video, what is HEAV? HEAVA is the vertical displacement of a vessel and its system of water caused by waste. And it's really important for many marine applications to compensate for this HEAVE to guarantee safer and more efficient operation. To do that, it's really important to have a sensor at MRU that provides accurate HIV, and this is why XSENSE MRU has been developed. We provide, indeed, real-time HIV data with an accuracy better than 5% up to 29 seconds HIV period and better than 6% upto 40 seconds HIF period. The sensor placement is flexible, it is not needed to place the sensor in the center of rotation or in the point of interest, but can be placed in any point on the vessel. And Ayush will talk and explain a little bit more in detail how this works. XSense Heave is available on our XSEN series and XSens AVR MRU. and if you already have an XS series or AV or AHRS, you can simply upgrade the firmware to the latest version for free and you will have access to the HIV output. I would like now to invite Ayush to join the floor. Hello all. Thank you, Martina. I'm Ayush, and I will now be taking all of you through how does the Heave filter work and what kind of performance can you expect from this filter. So as Martina explained, Heave is the vertical displacement of any vessel. And how do we obtain that, the ideal solution is to just take the local free z-axis acceleration and do a double integration of that. This acceleration comes from your HRS, which uses both accelerometer and gyro to find its orientation, and then also tell you what the local z- axis acceleration is. And if you just do a dual integration on that, you get your vertical displacement. but this is just the ideal case because in reality what happens is that your accelerometer also measures a certain bias along all three axes but here the z-axis is more relevant and if you just do a double integration without doing anything about this bias then in the output you have this quadratic term which scales quadratically with time so essentially you end up with a drift in your heave estimated position which varies with the square of time so that's not very ideal a naive solution to this problem is to just use a high pass filter which if tuned properly can get rid of this constant bias so you ensure that all the constant offset in your acceleration goes away and you're left with just the periodic component And if you double integrate that, you get your heave position quite nicely. A challenge with this approach, however, is that the bias drift and the phase performance of the heave estimate itself varies with the heve frequency or with the period heave period that you are encountering. so just having one cutoff frequency of this high pass filter and just one phase performance is not enough if you want a good accuracy so there is some real time adjustment needed and that's where we designed our heat period estimator as you can see on the slide so that uses the excel accelerometer signal to estimate first of all the heat period but then this heat period is also used to tune the high pass filter it adjusts the cutoff frequency and the phase behavior of the high pass filter and that ensures that you get real-time heave which is adjusted for the sea state urn and the kind of dynamics your ship experiences um now i would like to show you the performance of the heave algorithm we did several tests both inside our laboratory on the robots where we simulated different kind of motion profiles but also as you saw in the video we did some testing in the North Sea and I'll show the results from the North sea in a few slides but here something is a test that we did on the robot just we executed simple harmonic motion so what you see here the white curve is the ground truth so it was the position logged by the robot and the blue curve is heave that's estimated by our MRU and what we tried testing here is what happens when there's a sudden change in heat period and in practical situations something like this may occur when you are for instance offloading some weight off of your ship for example if it's hanging on a crane and if you offload the weight there's certain reduction in the ship's mass and that might cause a change in the ship's dynamics and that may adjust, that may affect the heave period of the vessel. So at 800 seconds, you can see that the heve period changes from 25 seconds to 20 seconds. And before that, the filter is converged and it takes about 50 seconds for the filter to estimate the correct period and reconverge its heave output. The performance, the error performance is more interesting, which you can see on this slide is just the error between the ground roots and the heave estimated by the MRU and you can see that before in the regions where the filter is converged we have less than five centimeters and five percent of error actually it's more close to two centimeters but this is just simple harmonic motion so it's quite easy to estimate. These are the results from what we had at North Sea. Again you see the ground truth data in the white curve on the top graph and you have the heave estimate from the MRU in orange. And in the subplot at the bottom you can see the error performance and again you can say that the error stays below the plus minus five centimeters of envelope. However there's one caveat here. You can see that at the beginning and somewhere in between with the arrows that I've marked there are regions of high dynamics so these are the regions where the ship was accelerating really at a really fast pace and what these high accelerations do is that they tend to degrade the orientation estimation of the sensor which propagates to the heave estimation but other than that in the regions of low dynamics you can see that the error performance is below the five centimeter bound that we claim. And that brings me to this range map of supported heave amplitudes and periods. So up to from 0 to 29 seconds we claim that our heave algorithm can estimate the heave with five centimeters or five percent of accuracy and if you go above that period if you are between 29 and 40 seconds the error is below six centimeters six percent and the higher you go the higher your period becomes the higher this error gets but 40 seconds is where we have measured the error to be maximum of six centimeters or six percent of course there's also a lower bound on this spec there's a region if you have very high frequency heave with very high amplitudes you will naturally cross the ag limit which our accelerometers are capable of measuring so there's of course a lower bound also on the heave estimation um finally there's the topic of heave estimations at offset points so we understand that it's not always possible to mount your sensor at your point of interest where you want to measure the heave so for that we allow the option of specifying lever arms to two points which is one of them is the center of rotation cor so this is the point at which all the effects of rotation on a vessel are cancelled out and the second point is the point of interest poi which is where you wants to measure your heave now of course you can mount your sensor at poi but it's better to mount it somewhere else and then specify these lever arms because what happens is the way the filter is designed if you have a constant tilt in your vessel for example due to wind then you will see a change in the heave output by the sensor but it will gradually fall off to zero because that's what the high pass filter does but if you mount the sensor somewhere else and you specify this lever arm specify both of these lever arms then you will see that even with constant tilts the heave output remains accurate with that i hand the floor back to martina if you have any technical questions please feel free to drop them in the questions tab and we will answer them shortly thank you Thank you, Ayush. Yeah, just last slide before the Q&A session, just wanted to mention that our product can be easily quickly tested through the development kit. Each development kit comes with the sensor, of course, and some accessories. Here you can see XS-AVIOR development kit with also the development board and the cable and XS series development kit, with the proper cables. It also comes with the empty software suite, empty manager for a quick configure and showing real-time data or recorded data, the empty firmware updater to update the firmware, magnetic fieldmapper and the software development kit with different programming languages. Extensive documentation, user manual, full integration manual including pin specification, drawing, 3D models, articles, tutorials, and many more. You can also contact support at movala.com to receive free technical support from our field application engineer. All right, we can now start the Q&A session. I will close the presentation. I see some questions already in the question tops. Let me see. So how is the accuracy of your MRU compared to Kongsberg MRU? So Kongsber MRU should be tactical, if not navigation grade MRU. So they are really bulky and expensive. Of course, they provide a higher, much higher accuracy. But our XSavir and Cirrus MRU are more compact suitable for also application where there is size constraint and still accurate. I would like to add to this answer. I believe we didn't benchmark directly against Kongsberg but we do believe that their spec is still five centimeters slash five percent and that's also what we have measured so we can say that we are still competitive in that regard yeah so you can find a similar accuracies if not the same in a more compact solution what is the heave latency well he latest is a real time so phenomena here should be like sub two milliseconds uh which can you control that yeah the algorithm itself is real time i believe any latency that is there should be because of the communication between the sensor but I'm not sure what the number is for a series and avia I think yeah should be really some milliseconds how are you calculating your heave accuracy figures less than five percent up to 90 29 seconds give period um so about centimeters that just subtracting your reference signal from the estimate the percentage is a bit tricky because when you divide a sinusoid by a sinus where you have these zero by zero regions which gives you problems so the five percent spec it comes from the testing that we did on our robot so it locks the absolute position in millimeters and we use that to calculate the percent uh error between the reference signal and the estimated signal and we did testing at different periods and also from the math from the filter design we know that at 29 seconds is where your filter response starts to flatten out a bit thank you ricardo is asking can we use this heave sensor on a train that's an interesting question uh well of course any of our product can be used in any application everything depends on what you need uh i i would recommend you to reach out to us so we can uh discuss and understand better your application your need so we can also direct you on the best solution but yeah i wouldn't say that cannot be used for uh on a train Patrick. Hi, Patrick. Will the heap feature also be available for other modules such as MTUI1 and 600 series? At the moment, we don't have any plan to extend these functionalities on other series. Extended AVRN series are our highest-end solution and also able, because of their design to properly handle any kind of vibration any kind of dynamics so we really want to guarantee the best performance on that but please feel free to reach out and yeah tell us what you need if you need if you like to have a different form factor we always listen to you to our customers over the market needs Another question, do you use a harmonic oscillator math model to aid the estimates of heave? Ayush, this is for you. No, we do not. We just use regular double integration and high-pass filters. And the reason for that is in the ocean, these math models, they don't really hold up because he states well the harmonics encountered in the ocean they are they have a much broader spectrum so if you just have a simple harmonic heave motion or your motion comprises of discrete frequencies then these models help but for a robust performance it's best to just stick to the simple stuff so that doesn't really help all that much Then Fabio is asking, have you already conducted a comparative test with the DataWell WaveRider WaveBoy, which is the reference for ocean wave measurements? Well, we haven't, but we are planning to extend our testing procedure, especially now that the sensor is live on the market, we can gather more data and refine, say, our testing. Patrick, is the HIF feature also useful for other non-maritime applications, for example, mobile mapping system to calculate the vertical fluctuation of the vehicle, for example due to uneven ground from the scan data in order to obtain a precise point cloud there's also a really interesting question are you sure you want to um to answer your question patrick i would say that if you are sure that your mean height is always zero yes it can be useful otherwise i would suggest that um you would use more uh the gnss and rtk versions because the way the filter is designed is that if you have it it assumes that you always have you are oscillating about a zero mean line so if you have some static displacements then your filter will show a displacement response and then it will fall off to zero so that's for heave that's okay but for absolute altitude cracking that's not what you want. So if you are sure you are just oscillating about one mean line then it's okay, otherwise it's best to use the GNSS version of our portfolio. Question from Dan. How long is required for this sensor to initialize the heave reference point? Is this configurable? i'm not sure i understand what you mean by initializing the reference point you can specify the lever arms in empty manager so when your sensor is connected you can specify those both the leverarms for a center of rotation and point of interest and effective immediately once you press the apply button these lever arms change in the device and the heave output then is the heave output for the point of interest now of course it does take 50 seconds of convergence if you start heaving immediately then you will have some time to converge 40 to 50 seconds but after that you track heave real time then thomas is asking are the heav estimates Oh, sorry. I think we missed the question from Jack. Are you allowed to state the order of the adaptive high-pass filter? No, that cannot be altered, no. Yeah, I missed that one. As in, you cannot manually alter it over empty manager, but the filter itself, it does quite some adaptations to ensure that the phase lag induced is minimal. Thomas, are the HEAV estimates available as objects on the CAN bus? Currently we are working on a stable release for this. We provided for now an alpha release so some customers can test the HEV output through CAN. It will be available soon. Question from Ranga for Ayush. Could you please re-explain why an adaptive cutoff frequency must be set for the high-pass filter? Well, the high pass filter is not just adapting its cutoff frequencies, it's also adjusting its phase response. But the main reason is that we observed is that if you just have one cutoff frequency for the high pass filter you well you can leave at different frequencies let's just say and you need to choose the cutoff frequency such that it aligns more or less closely with what frequency you're heaving at so that you are pretty sure everything below that is bias and it's removed so the adaptations that are made are made to make sure that your output matches is what the actual heap motion is. All right. Another question from Dominique. Did you add some standard telegram in order to output the heap data into standard acquisition software? And that's more about the communication protocols and express messages, I think. Yeah. Well, we do support Eve output over all different communication channels. So SPI as per C can, and it can be read using those channels, but also we have our example codes so you can use c plus plus and with python to get these messages from our trackers another question from dominic do you have a quality indicator to give the start of heath calculation anytime needed at startup to initialize the printer if you mean by quality the uncertainty no that's not something that we support at the moment if you send us feedback that's something that we can look into for the future any time needed at the startup to initialize the filter no it is running as soon as you connect the sensor but yes if you change the heat frequencies and the filter does take around 50 seconds to converge to the new heave estimate but from our experience in the ocean the changes are not be as abrupt so the periods vary over long time scales and the heave output is more or less tracking the real motion we have another question for you from neil is your sample frequency uh used to adopt the filter happening all the time no it happens uh all the time. The adjustments are taking place at the same time as the heave filter is running, so that's 100 hertz which is the rate at which we output heave and also the rate at which the computations are performed. Harald is asking how do you estimate sorry that's a that's that's confidential that's an algorithm that we've worked on ricardo high frequency this can be measured what is the highest frequency i believe that you can go up to 10 hertz sorry not 10 hertz 5 hertz but we can also of course increase that range but we had a practical consideration in mind that no ship will be heaving at 5 hertz so we set a limit of 5 hertz all right i don't see any other question yes there is another question why not a bandpass filter instead of a high-pass filter then the noise can be filtered out without getting the noise also integrated that's a good point diranga but the double integrated itself it already attenuates the noise quite well it has a very low gain and high frequencies so any added bandpass filtering does not add much advantage and with the bandpass filter you also run the risk of attenuating the frequencies at your point of at your frequency of interest so you really need to be careful with the shaping of the band pass filter bandwidth so in our performance tests just a high double integrator and the high pass filter works quite well one more question am i right to assume that the use of this sensor for heat compensation slash control of an item that is lowers of the ship would not be a fit for this the top so you mean i think yeah probably like system underneath thinking of the ship i'm not sure i know so emily feel free to reach out to us so we can understand a little bit better your question and we can for sure support you All right, I hope we answered all the questions. If some are unanswered, we will reach out to you or again, feel free to contact us. We're happy to start a conversation and I would like to thank everybody and also Ayush. I hope you enjoy the webinar and thanks also for the really interesting question and have a nice rest of the day. Thank you.