Hi, Consider a scenario where you are predicting some state

x

at time

t

based on historical state measurements of

x

. At any given time

t

, the predictor only has access to

x

measurements up until some time point

t_latest_meas(t) < t

due to delays. We cannot assume that

t-t_meas

is equal to some constant time delay

D

in this scenario. We can log

t_latest_meas(t)

every time we log a value for

t

. Let's call the predictions

x_pred

and the measurements

x_meas

. In scenarios like this, I would find it useful to visualize in a time series view the evolution of the

x_pred

along with the history of

x_meas

available to the predictor at any given time. To clarify, I want to create a time series view in rerun with the following properties: - When the timeline cursor position is set to

T

, I want to show

x_pred

for all

t <= T

- When the timeline cursor position is set to

T

, I want to show

x_meas

for all

t <= t_latest_meas(t)

I've also attached a video showing what I'm going for. In this video, the blue curve represents

x_meas

, the red curve represents

x_pred

and the black bar represents the cursor position on the timeline. (Here,

x_pred(t)

is simply taken to be linearly extrapolated from the speed of

x_meas

from some fixed time delay back). Now, my question is: Can I achieve this in Rerun, and if so, how? The closest I can think of is to set different

VisibleTimeRange

for individual entities (like shown [here](https://github.com/rerun-io/rerun/issues/8557) ), but then I don't think I would be able to dynamically update the upper bound of the visible time range of the

x_meas

entity to

t_latest_meas(t)

at every logged prediction

x_pred(t)

, unless I resend the entire blueprint every time. I haven't actually tried this, but it doesn't seem like a feasible solution. Thank you https://cdn.discordapp.com/attachments/1534219749146234991/1534219749469192383/Animation.webp?ex=6a7354fc&is=6a72037c&hm=c1f54165baa1ca31f676cdd999f08cc32ea7b8e71b6baa9fe21faebb7ed19df2&

Hello everyone! I have a quick question about your job openings. I recently applied for the [Software Engineer, Infrastructure](https://rerun.io/careers/a6b9d129-a0fc-4635-96ca-8967c8617808) position. The description says that the team has members in European and US East Coast time zones, with everyone available for synchronous work between 1:30 and 5:00 p.m. CET. However, I later noticed that the role is listed as “Remote · East Coast, US.” I’m based in Europe—would my location still be acceptable for this position? If not, I would also be interested in the Backend position, as I have a C++ background too. I really like what you’re building and wouldn’t want to miss the opportunity because of a misunderstanding about the location. Thanks a lot! Jan

Hi - I am using rerun to log data from my robot system. one thing I am logging is pose. I Want to be able to view this pose on a outline of our map. I have created a separate rrd file with the map outlin in it, set

static=true

and the same application id as my poses logs . When I run the viewer with something like

rerun map.rrd poses.rrd

I get 2 separate recordings in the viewer. If i override the recording id in both files to be the same, then I get one recording in the viewer and can view the poses on the map. This means I cannot use the recording ID to differentiate recordings if it has to be the same as my statis map file... Is the only way to achieve what I want to actually log the map data each time I collect pose data so it is in the same log (and recording id) as the poses? Thanks!

I work on space and geospatial applications and would like to be able to use double precision to represent positions. I can sort of fake this by having independent coordinate frames I log in separately, but this is less elegant than if Rerun had the ability to have native double-precision transforms. Is this something that you've thought about?

Hi everyone, I am working on a C++ application that involves visualizing multiple FFT magnitude spectrums (magnitude vs. frequency) in real-time alongside my other time-based data. I'm looking for advice on the best way to log and visualize this type of data. Here is what I have tried so far: - **Scalar (with a frequency timeline)**: This doesn't allow me to visualize the spectrum in sync with my other time-dependent data. - **LineStrips2D**: This somewhat works, but it is very difficult to visualize and interpret in the viewer. - **BarChart**: The x-axis defaults to bins rather than actual frequency values, which isn't ideal. I saw a post that accomplished something very similar to what I want (specifically the middle view): https://discord.com/channels/1062300748202921994/1062300748777529381/1519662422644883546 If anyone has any information or code snippets on how to accomplish this, it would be greatly appreciated! https://cdn.discordapp.com/attachments/1526753414162157609/1526753414338580510/image.png?ex=6a582b6b&is=6a56d9eb&hm=0623b9a6a0b83b9778e0fd6bed4cde8ad5d59f30ee76efffe7ca1cd02624acc9&

Hey there ! I'm trying to replicate [this example](https://rerun.io/docs/reference/types/archetypes/scalars#multiple-time-series-plots) using the Python SDK, but each plot is shown on a separate view. In particular

trig/sin

and

trig/cos

are not plot on the same view despite the shared root. Is there something I'm missing ? 🙁

python3.12
rerun-sdk==0.33.0

https://cdn.discordapp.com/attachments/1521095529193537586/1521095529398931486/Screenshot_2026-06-29_at_12.10.29.png?ex=6a43961a&is=6a42449a&hm=d8d137d7fc79aa5e57d59bfad76daed02c88c5e38355d1efa45179496dd47f46&

To animate my robot over time, so far I've tried two things: 1) a for-loop, calling

rec_.log('transforms', rerun::Transformd3D())

2) using

send_columns

with

rerun::Transform3D().with_many_*().columns()

I imagine the intended method is 2) but it has the downside of me having to specify all timelines, and I have to create a copied-vector like so:

// all transforms here are logged with the same timestamp
  std::vector<int64_t> step_repeated(transforms.size(), log_step_);
  const auto rr_indices = rerun::TimeColumn::from_sequence("step", step_repeated);
  rec_.send_columns("transforms", rr_indices, rr_transforms);

How come there isn't an option to just

rec_.log

with a Transform3D that has multiple transforms in it? Like you can for many other things (e.g. Points3D, LineStrips3D, ...) Thanks!

I frequently find myself wishing there was a good way to use rerun to compare the outputs of two experiments visually. Sometimes it's like scalars, but in general it could be things like robot motions. Rerun is great at visualizing one experiment's data, but what about two? In general I would like to easily choose the experiments I've already logged, and have their views overlayed/combined in some way. For scalars it's pretty clear, you just put them on the same plot. For 3D or 2D views I think you can again simply add them to the view, and then I can use alpha/opacity or draw-order, or just toggle on/off different ones with the visiblity icon. One could log separate recordings and click between them, but that's clumsy and imprecise. As a reference, for scalars, PlotJuggler handles this great. I can open two CSV files, and it use the file name as a prefix, I then have all the data from both files that I can drag into whatever views I want. So maybe rerun can do something similar? where I would load two rrd files, and it would prefix each entity with the rrd filename (or let me customize the prefix at open/load time, like PlotJuggler does)? So is anyone using rerun for stuff like this? Is there a good workflow for this? Is anything related to this on the roadmap?

I’m using Rerun 0.32.0 on Ubuntu 22 to visualize ROS2 data from a Python subscriber/logger. I log data from multiple lidars and cameras. After starting the Rerun viewer and running my Python script, the viewer sporadically crashes after some time with: thread 'rayon-12' panicked at 'internal error: entered unreachable code', re_video/src/demux/mod.rs:858 The crash does not happen immediately, but only intermittently after the system has been running for a while. Does anyone have tips on how to narrow down the root cause?

Hey there, I'm just hoping for guidance on an issue I've been dealing with when trying to embed an instance of rerun into a simple egui app. I have an instance of the rerun Python SDK running on a host (let's say 10.0.0.1), which is doing

rr.serve_grpc()

. This instance is also logging some test text log data for me to see. When I run

rerun --connect rerun+http://10.0.0.1:9876/proxy

, the viewer opens and I see the logs as intended. Now I have the re_viewer embedded in a tiny egui app based on the [extend_viewer_ui example](https://github.com/rerun-io/rerun/tree/main/examples/rust/extend_viewer_ui), with a code snippet as follows:

rust
// in an egui ui() function...
if my_button_clicked {
    // connect rerun to the selected host's server using its ip address
    if let Some(host_name) = &self.active_robot {
        let robot = &self.robots[host_name];
        info!("Connecting to {} at {}", host_name, robot.ip);
        
        let url = format!(
            "rerun+http://{}:9876/proxy",
            robot.ip
        ); 

        // ^this evaluates to rerun+http://10.0.0.1:9876/proxy

        self.rerun_app.open_url_or_file(&url);
    }
}

When I click the button, in the console, I see

log
[2026-05-10T08:36:03Z INFO  my_app] Connecting to TEST at 10.0.0.1
[2026-05-10T08:36:03Z DEBUG re_viewer_context::open_url] Opening URL: RedapProxy(rerun+http://10.0.0.1:9876/proxy)
[2026-05-10T08:36:03Z DEBUG re_grpc_client::read] Loading via gRPC… uri=ProxyUri { origin: Origin { scheme: RerunHttp, host: Ipv4(10.0.0.1), port: 9876 } }
[2026-05-10T08:36:03Z DEBUG re_viewer::app] Adding new log receiver: rerun+http://10.0.0.1:9876/proxy

...but no signs of life in the actual viewer UI. I'm not quite sure why the standalone and embedded versions are behaving so differently. Any help would be appreciated, I'd really like to use rerun! :)

Hi there, I am doing an experiment where I save it to a .dat file then I launch the viewer: uv run rerun rerun.dat [2026-05-07T18:08:14Z INFO winit::platform_impl::linux::x11::window] Guessed window scale factor: 1 MESA-INTEL: warning: Performance support disabled, consider sysctl dev.i915.perf_stream_paranoid=0 [2026-05-07T18:08:14Z INFO re_grpc_server] Listening for gRPC connections on 0.0.0.0:9876. Connect by running

rerun --connect rerun+http://127.0.0.1:9876/proxy

[2026-05-07T18:08:14Z INFO re_data_loader::load_file] Loading "rerun.dat"… [2026-05-07T18:08:14Z ERROR re_viewer::app] Failed to open data source: "rerun.dat": No data-loader support for "rerun.dat"

I have a behavior tree which I would like to render in rerun, so I can view it next to the 3d robot view and scrub back and forth in time etc. I have tried a couple approaches 1) The native graph representation. It ended up being quite difficult to see the information because of the automatic node placement. (edit: I did try manual placement as well) 2) Rendering lines and text in a Spatial2D view. This somewhat solved the node placement problem but I couldn't specify the font size in the Spatial view context so the text would overlap if there were too many nodes. If I could specify text size relative to the Spatial 2D view this approach should work. 3) Rendering the tree as text. So far the text approach is the closest but, there are a couple minor inconsistencies. I would like to be able to render lines, and use bold strike-through and other unicode characters to held distinguish different types of node. The problem is that none of the text modes I have tried support all the required Unicode characters, each different mode supports a subset. (see attached image). The hybrid mode on the right technically can do everything required, but the lines are highlighed because of the inline code hack to get working | symbols Is there a way to work around these issues? Thanks https://cdn.discordapp.com/attachments/1500911768048762970/1500911768376184832/image.png?ex=69fad143&is=69f97fc3&hm=b759b1dd42edd858f98b52edbb3522ca0083885ac10d8f97a9b9a5c84bd22744&

Hey I wanted to know what the status for lazy loading .rrd files is for the main client, I know there's been some working going on the oss server side to make it possible. I have very large rrds ~50gig's in size that I wanna view on 5g-10g memory budget, I was able to implement a patch that uses re_server (before all the new patches), to make it work with lazy loading was just wondering if it's in the plans to also make it work for the main client.

Hi, I can load 2 RRD files using rerun file1.rrd file2.rrd They both appear in the sources panel and I can select one or the other. Is there a way to combine the data from 2 files in the same viewport windows? I have set the recording_id to the same on both files. Does the application_id have to be the same also? Thanks

Hi, why does the Rerun only get white when I try to do lerobot-record --robot.type=so101_follower --robot.port=COM10 --robot.id=my_awesome_follower_arm --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 1920, height: 1080, fps: 30}}" --teleop.type=so101_leader --teleop.port=COM11 --teleop.id=my_awesome_leader_arm --display_data=true --dataset.repo_id=JFI123/record-test --dataset.num_episodes=5 --dataset.single_task="Grab the black cube" --dataset.streaming_encoding=true --dataset.encoder_threads=2 https://cdn.discordapp.com/attachments/1486412593395667114/1486412593710235648/image.png?ex=69c5691a&is=69c4179a&hm=6af589f2ac3c68eb2d0eff72d5032f41f5c0a3e4a116080fffa4d2163f7036fe&

Is there a good way to log binary data of some form, e.g. network packets? I considered using the file logging API, but I don't know if there are any data loaders supporting raw bytes, and I don't think treating them as files is the most appropriate either.

Hey! I'm logging lane and object detection predictions in rerun. Each have their own confidence values which I'm currently logging as labels. We also have a hardcoded thresholding for the confidence, so anything below that confidence won't be logged, and anything above will. This works okay, but what would be really useful is if we could have some sort of "slider" that allows us to view or hide objects at view time based on the confidence values. Is there a way to do this? This would be especially helpful because as the model learns the confidences increase, so a hardcoded thresholding doesn't work so well when viewing at difference stages of learning. Thanks!

Hey! I'm using explicit transform frames as described in the Transforms & Coordinate Frames docs to decouple my spatial relationships from entity paths. I hit what seems like a bug when I log a

Pinhole

and a

CoordinateFrame

on the same entity, the image disappears from 3D views. Removing the

CoordinateFrame

fixes it. The

CoordinateFrame

docs say you can associate entities with named transform frames, and the Pinhole docs mention that Pinhole projections define 3D-to-2D spatial relationships. I'd expect these to be orthogonal or independent —

CoordinateFrame

just declares which frame the entity belongs to, it shouldn't affect whether

Pinhole/Image

data is found.

rust
use ndarray::{Array, ShapeBuilder as _};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let rec = rerun::RecordingStreamBuilder::new("pinhole_coordinate_frame_bug").spawn()?;

    let mut image = Array::<u8, _>::default((100, 100, 3).f());
    for ((y, x, c), val) in image.indexed_iter_mut() {
        *val = match c {
            0 => (x * 255 / 100) as u8,
            1 => (y * 255 / 100) as u8,
            _ => 128,
        };
    }

    rec.log_static(
        "world",
        &rerun::Transform3D::default()
            .with_child_frame("camera_optical")
            .with_translation([0.0, 0.0, 1.0]),
    )?;

    // Camera A: Pinhole + Image, NO CoordinateFrame — renders correctly
    rec.log(
        "world/image",
        &rerun::Pinhole::from_focal_length_and_resolution([100., 100.], [100., 100.]),
    )?;
    rec.log(
        "world/image",
        &rerun::Image::from_color_model_and_tensor(rerun::ColorModel::RGB, image.clone())?,
    )?;

    // Camera B: Pinhole + Image + CoordinateFrame — image is invisible
    rec.log(
        "world/image_with_frame",
        &rerun::Pinhole::from_focal_length_and_resolution([100., 100.], [100., 100.]),
    )?;
    rec.log(
        "world/image_with_frame",
        &rerun::CoordinateFrame::new("camera_optical"),
    )?;
    rec.log(
        "world/image_with_frame",
        &rerun::Image::from_color_model_and_tensor(rerun::ColorModel::RGB, image)?,
    )?;

    println!("'world/image'            — no CoordinateFrame — should render");
    println!("'world/image_with_frame' — with CoordinateFrame — image is invisible");

    std::thread::sleep(std::time::Duration::from_secs(60));
    Ok(())
}

world/image

renders,

world/image_with_frame

has errors. The only difference is the

CoordinateFrame

. Is this a known issue? Am I doing something wrong? https://cdn.discordapp.com/attachments/1481382681483673691/1481382688479907861/2026-03-11_at_20.04.472x.png?ex=69b31ca3&is=69b1cb23&hm=69bb54d8657797e9541928d0672573ae99c50c783ca9577e9e79592ddc1ed4a0&

hey, I'm trying to visualize my sensor's data. I recorded it for 45 mins and its nearly 500 Hz. I want to plot 7-8 scalar graphs. But as u see below my CPU is like that. when I examine my data graph in rerun viewer. what is the problem and how to fix it, could u help me please? https://cdn.discordapp.com/attachments/1480909072944336919/1480909073141465119/image.png?ex=69b1638c&is=69b0120c&hm=60f8d249dcb14e2296251a8967689ae15cbabcf47fdea82e868e7db504fbcf08&

Hello there, this is an egui question, I did not find a dedicated Discord, I apologize if it is not the right place. Some context I work for an Air Navigation Service Provider, we provide Air Traffic Control software for countries in the world (like Thalès) and we are doing a POC on whether egui is a good fit to create a new radar HMI. The interface includes classic toolbars, floating windows, and different control menus, but the main part is a map (using Walkers) on which we project mainly two things: - Geojson based graphics and sprites representing areas like airways, restricted airspace areas and so on. - Track (aircraft) elements: - a symbol representing the type of the aircraft - a rect surface on which we display different texts like ground speed, aircraft callsign, flight level etc... - a vector line linking the symbol to the label - a vector line starting from the symbol and drawn up to the expected aircraft position in x minutes - a trail of past positions represented by e.g. dots or circles We aim to support more than 2k simultaneous aircrafts which could eventually be all visible at the same time if the air traffic controller "dezooms". As a result it would account for: - 2 000 symbol svg images (10 possibilities) - 6 000 texts - 4 000 vector lines - 10 000 vector circles or dots + all the areas drown as graphics behind which could account for 10 000 additional vectors in rare cases The radar update rate through a websocket is approximately one message containing a batch of 100 tracks every 200 ms. Every individual element of a track is interactive and display different color on hover independently. We conducted some tests with both web and native targets and are thrilled to see it does seem to handle it, but do you have any recommendations to share and caveats we should be aware of? Since this is for safety critical software, do you have any counter indication? I hope you find it as cool as I do, egui for ATC would be 🔥 https://cdn.discordapp.com/attachments/1480553783997239398/1480553784299094057/Screenshot_2026-03-09_110650.png?ex=69b018a9&is=69aec729&hm=4cf268f87ac47427fcd2ef22dcf3404aa3b309ac7c6f1667ed3f2fb4f1dd2e8f&

Hi, I want to access the view data by directly accessing view buffer or taking screenshot. I know of the experimental screenshot API, albeit it hasn't worked for me yet. I run rerun in a headless remote server which seems to magically launch a local viewer in my system. I wonder what options I have to access the data in the viewer?

Hi there. This is my second quetion here. But let me reintroducing my goal: I am considering using Rerun for my robot visualisation/debugging Tool. I am currently developping this under dash framework but this is tedious. I have hovewer some use cases I don’t know how to solve in rerun. For instance: rerun-sdk support pinhol camera model. But AFAIK not cylindrical camera models, Am I right? What would be the preferred approach ? To create a PR to rerun repository to have this available there? Or to transform all my 3d points so that it get projected correctly using the pinhol model? Best is that you have the implementation already somewhere 🙂 Thanks for your support! And for the awesome Rerun platform!

We embed the Rerun web viewer in an application that uses a different color scheme (dark navy-blue backgrounds, purple accents). The viewer's default theme visually clashes with our host application. Is there an existing way to customize the viewer's theme at runtime that we're missing? We looked at the JS API options and CLI flags but couldn't find anything. If not — it would be great to have some way to override the theme without forking and rebuilding from source.

Hi there I am considering using Rerun for my robot visualisation/debugging Tool. I am currently developping this under dash framework but this is tedious. I have hovewer some use cases I don’t know how to solve in rerun. For instance, at time T, I want to plot a 2D scatter of the values of one particular 2D array. At time T+1 this array has different values that would then overwrite the old plot. Is that possible? How? Thanks for your support! And for the awesome Rerun platform!

Hi everyone, I'm wanting to distribute a tool as a uv package and bundle the rerun viewer along with it. I've depended on the SDK with

[project]
...
dependencies = [
    "rerun-sdk>=0.29.0"
]

Which could then be installed using

uv tool install

But that still requires users to manually pip install the SDK to get access to the viewer, does anyone know how to bundle the viewer with the package to avoid this, and to keep the SDK dep here at the same version as the viewer?

Hey, I am new to using rerun, and I am using it to visualize videos of thermal cameras. I have detection masks for which I have logged the average temperature over time, but I would also like to explore part of the images interactively where currently no detections are made. Currently I am co-opting the DepthImage archetype for this, because it allows floating point images and on hover shows exact pixel values, which is already useful. However, I would like to add some features and I'm not sure where to begin. On hover, I would like to see some statistics for the selected region, such as the mean value of the region or even a histogram of the depth values. Even better would be if I can select a point or region in the depth image, and visualize the depth over time. Does anyone have an idea if something similar already exists, or where to start with implementing it? https://cdn.discordapp.com/attachments/1472996198544572728/1472996199278706732/image.png?ex=69949a1c&is=6993489c&hm=d418d719b18327e23b44fa0477406e7171191458f206b3de5336b22602baf0b3&

I can see the demo images on the rerun website have no color over non masked areas but I havent been able to get that working. I've set the value to zero and tried setting the colors of other classes in annotation context but neither have worked for me.

I am doing keypoint annotation for hands in egocentric data. I have used SAM3D body, but the annotaion is not overlapping in the video. can I edit that in rerun?

I have a use case in mind where 2 or more source files get converted to an RRD. Do I need to take care to merge the source data in time order when logging to RRD, or can I log one set of source data, jump back in time and log the 2nd set, and the rerun viewer will sort out the time ordering? Thanks

When running compaction is it safe to do it inplace, with

⁨rerun rrd compact file.rrd --output file.rrd⁩

, or is there a chance of the file being overwritten while being read?