My company put out a request for sample datasets that include IMU (or just accel) measurements. I thought folks may find the post relevant and could help us out. See the announcement below and let me know if you have any questions. I know they are still handing out gifts. Thanks in advance.
We’re testing an online IMU data processing utility and need a set of sample datasets to test against. There is no standard file format for IMU sensors like accelerometers and gyroscopes. The formats (csv, txt, xlsx, mat, bin, hdf5…) units (G, m/s^2, rad…) and data types (int, float…) all vary. Hence this post.
Do you have a dataset that includes IMU (or just accel) data which you could share? We can use it “as-is”. As a small “thanks” we’re sending $15 Amazon gift cards (1/person while supplies last). Here is the site to share data: Easily Process IMU Data Online . Two steps:
Select file(s)
Please do NOT clean, trim, or fix the dataset. We want to keep it realistic.
A “readme.txt” with a little context is nice by not required.
Since you don’t specify a format and say “as-is” is OK, I’m assuming that bag files with IMU data are of interest to you? I’d imagine that’s the format most people here would be able to provide you with.
@navhm Actually it is about data format, bitwidth and stamps. I think, now for sharing it will be the best way to record and share bag files containing IMU messages with any kind of ground truth (odometry, or /cmd_vel messages, for example). What do you think?
Strictly speaking, yes. Anyone can define his own message types and still save them into a BAG file. However, 99% (qualified guess ) of ROS users store sensor_msgs/Imu messages to a BAG file when recording IMU data. So here, the data format and file format are kind of the same.
@peci1 - thanks for pointing out the documentation and recent PR proposal. I’ve worked with a number of IMU sensors and will take a look. As you stated, the beauty of frameworks like ROS is that most folks use the standard messages and that can abstract away the underlying manufacturer. To @twdragon 's point - we’ll see how true that is from the sample sets received.
The other audience may be folks who are not ROS users (or other similar frameworks). They may use manufacturer formats or their own definitions for logging data (think IoT device or custom embedded system). Where possible, we want the utility to be useful to them too. Whomever works with IMU sensors and could use an online utility for managing and processing data.
BAG files have come in. There must be many BAG files collected over the life of a project. This may be an odd question, but what do folks do with their BAG files? Is there ever a reason to revisit them (e.g. from 6 months ago)?
Sure! For research, we usually record all data during experiments, even those not vital for that particular case. We often later find that such datasets are useful for some other research conducted later. Also, you can generate various visualizations using Bag files, which may be useful when you’re making some presentations or PR output. Last, if you e.g. improve your localization or mapping algorithm, you can later test it on lots of recorded data without the need to physically revisit the places.
The bags are very useful because you can completely reproduce the environment and command structure in which the robot operates.
BTW, @navhm, can you share some info about your company? It could be very interesting to know the new professional participant of the community of robotics. Maybe, also your chiefs would be happy to announce their work on the dedicated post here?
@twdragon - yes the original post photo has our company name. Briefly, we develop (and customize) sensor fusion algorithms for different commercial applications, including robotics. This we do for client projects as well as specific services we offer ourselves. If interested, see our site “Services” page or reach out. I’ll keep this description brief to keep the thread on topic. As you mentioned, maybe a dedicated post could make sense in the future.
If you’re searching for data from a MTi-10, here is a dataset of over 18.8 km of autonomous Teach-and-Repeat on cross-country ski trails in the boreal forest.
As part of our initial review of the datasets we’ve compiled a summary of stats and a few interesting notes (see below). Folks here may find this interesting.
We’d put out this data request in several places. By the end of the month we had collected several dozen submissions - enough for service testing prior to the soft launch (sign-up here to get notified of service updates: (Soon) Online IMU Data Processing ). The ROS community was most generous. Thanks for taking time to comment, read, or share data.
Given that ROS is becoming the industry standard in robotics, I thinks its interfaces could also be treated as standards =) However, it is already clear that the current interface is not good for all cases. That’s why 