User Interaction in WebGL (updated)

Duncan Murdoch

March 05, 2024


This document describes how to embed rgl scenes in HTML documents and use embedded Javascript to control a WebGL display in an HTML document. For more general information about rgl, see rgl Overview.

We assume that the HTML document is produced from R markdown source using knitr or rmarkdown. This format mixes text with Markdown markup with chunks of R code. There is a limited amount of discussion of other methods.

There are two ways to embed an rgl scene in the document. The newest one is recommended: call setupKnitr with argument autoprint = TRUE early in the document. This will set things up to be quite similar to the way standard 2D graphics are included by knitr, i.e. it will detect the fact that you’ve drawn something, and just include it automatically.

If autoprint = FALSE is used or no call is made to setupKnitr(), an explicit call to rglwidget will produce a “widget” which can be embedded into your document by printing it. This document uses that method.

Older methods (e.g. writeWebGL or various hooks) that were used before rgl version 0.102.0 are no longer supported.

Browser support

Most browsers now support WebGL, but in some browsers it may be disabled by default. See for help on a number of different browsers.


We start with a simple plot of the iris data. We insert a code chunk and call the rglwidget function with optional argument elementId. This allows later Javascript code to refer to the image. We also save the object ids from the plot, so that they can be manipulated later. (The first example in Controls uses tags instead of saving the ids.)

plotids <- with(iris, plot3d(Sepal.Length, Sepal.Width, Petal.Length, 
                  type="s", col=as.numeric(Species)))
rglwidget(elementId = "plot3drgl")

Next we insert a button to toggle the display of the data.

toggleWidget(sceneId = "plot3drgl", ids = plotids["data"], label = "Data")

The sceneId is the same as the elementId we used in rglwidget(), the ids are the object ids of the objects that we’d like to toggle, and the label is the label shown on the button. To find the names in the plotids variable, apply names() or unclass():

## [1] "data" "axes" "xlab" "ylab" "zlab"
## data axes xlab ylab zlab 
##   13   14   15   16   17

Using magrittr or base pipes

It can be error-prone to set the elementId in the rglwidget() to match the sceneId in the toggleWidget() (or playwidget(), described below). In the usual case where both are intended to appear together, magrittr-style pipes can be used quite flexibly: the first argument of the control widget accepts the result of rglwidget() (or other control widgets), and the controllers argument of rglwidget() accepts control widgets. In R 4.1.0, the new base pipe operator |> should be usable in the same way.

For example,

rglwidget() %>%
toggleWidget(ids = plotids["data"], label = "Data")

If you have R 4.1.0 or greater, this should do the same:

rglwidget() |>
toggleWidget(ids = plotids["data"], label = "Data")

You can swap the order of button and scene; use the magrittr dot (or the => syntax in base pipes) to pass the toggleWidget to rglwidget in the controllers argument:

toggleWidget(NA, ids = plotids["data"], label = "Data") %>%
rglwidget(controllers = .) 

or using R 4.1.0 or later,

toggleWidget(NA, ids = plotids["data"], label = "Data") |> 
  w => rglwidget(controllers = w) 


We have seen how to change the contents of the plot using toggleWidget. We can do more elaborate displays. For example, we can redo the previous plot, but with the three species as separate “spheres” objects and buttons to toggle them:

clear3d() # Remove the earlier display

with(subset(iris, Species == "setosa"), 
     spheres3d(Sepal.Length, Sepal.Width, Petal.Length, 
                  radius = 0.211,
                  tag = "setosa"))
with(subset(iris, Species == "versicolor"), 
     spheres3d(Sepal.Length, Sepal.Width, Petal.Length, 
     	       radius = 0.211, 
     	       tag = "versicolor"))
with(subset(iris, Species == "virginica"), 
     spheres3d(Sepal.Length, Sepal.Width, Petal.Length, 
     	       radius = 0.211,
     	       tag = "virginica"))
decorate3d(tag = "axes")
rglwidget() %>%
toggleWidget(tags = "setosa") %>%
toggleWidget(tags = "versicolor") %>%
toggleWidget(tags = "virginica") %>%
toggleWidget(tags = "axes") %>% 
asRow(last = 4)

Since we skipped the label argument, the buttons are labelled with the values of the tags. The asRow function is discussed below.

toggleWidget() is actually a convenient wrapper for two functions: playwidget and subsetControl. playwidget() adds the button to the web page (and can also add sliders, do animations, etc.), while subsetControl() chooses a subset of objects to display.


For a more general example, we could use a slider to select several subsets of the data in the iris display. For example,

rglwidget() %>%
playwidget(start = 0, stop = 3, interval = 1,
	   subsetControl(1, subsets = list(
	   			 Setosa = tagged3d("setosa"),
	   			 Versicolor = tagged3d("versicolor"),
	   			 Virginica = tagged3d("virginica"),
	   			 All = tagged3d(c("setosa", "versicolor", "virginica"))

There are several other “control” functions.


par3dinterpControl approximates the result of par3dinterp.

For example, the following code (similar to the play3d example) rotates the scene in a complex way.

M <- r3dDefaults$userMatrix
fn <- par3dinterp(time = (0:2)*0.75, userMatrix = list(M,
                                      rotate3d(M, pi/2, 1, 0, 0),
                                      rotate3d(M, pi/2, 0, 1, 0)) )
rglwidget() %>%
playwidget(par3dinterpControl(fn, 0, 3, steps=15),
 	   step = 0.01, loop = TRUE, rate = 0.5)

Some things to note: The generated Javascript slider has 300 increments, so that motion appears smooth. However, storing 300 userMatrix values would take up a lot of space, so we use interpolation in the Javascript code. However, the Javascript code can only do linear interpolation, not the more complex spline-based SO(3) interpolation done by par3dinterp. Because of this, we need to output 15 steps from par3dinterpControl so that the distortions of linear interpolation are not visible.


propertyControl is a more general function to set the value of properties of the scene. Currently most properties are supported, but use does require knowledge of the internal implementation.


clipplaneControl allows the user to control the location of a clipping plane by moving a slider.


Less general than propertyControl is vertexControl. This function sets attributes of individual vertices in a scene. For example, to set the x-coordinate of the closest point in the setosa group, and modify its colour from black to white,

setosavals <- subset(iris, Species == "setosa")
which <- which.min(setosavals$Sepal.Width)
init <- setosavals$Sepal.Length[which]
rglwidget() %>%
  vertexControl(values = matrix(c(init,   0,   0,   0, 
                                     8,   1,   1,   1), 
                                nrow = 2, byrow = TRUE),
                attributes = c("x", "red", "green", "blue"),
                vertices = which, tag = "setosa"),
	step = 0.01)


A related function is ageControl, though it uses a very different specification of the attributes. It is used when the slider controls the “age” of the scene, and attributes of vertices change with their age.

To illustrate we will show a point moving along a curve. We give two ageControl calls in a list; the first one controls the colour of the trail, the second controls the position of the point:

time <- 0:500
xyz <- cbind(cos(time/20), sin(time/10), time)
lineid <- plot3d(xyz, type="l", col = "black")["data"]
sphereid <- spheres3d(xyz[1, , drop=FALSE], radius = 8, col = "red")
rglwidget() %>%
  ageControl(births = time, ages = c(0, 0, 50),
		colors = c("gray", "red", "gray"), objids = lineid),
  ageControl(births = 0, ages = time,
		vertices = xyz, objids = sphereid)),
  start = 0, stop = max(time) + 20, rate = 50,
  components = c("Reverse", "Play", "Slower", "Faster",
                 "Reset", "Slider", "Label"),
  loop = TRUE)


While not exactly a control in the sense of the other functions in this section, the rglMouse function is used to add an HTML control to a display to allow the user to select the mouse mode.

For example, the display below initially allows selection of particular points, but the mouse mode may be changed to let the user rotate the display for a another view of the scene.

# This example requires the crosstalk package
# We skip it if crosstalk is not available. 

ids <- with(iris, plot3d(Sepal.Length, Sepal.Width, Petal.Length, 
                  type="s", col=as.numeric(Species)))
par3d(mouseMode = "selecting")
rglwidget(shared = rglShared(ids["data"])) %>% 

The rglShared() call used here is described below.

Layout of the display

Many rgl displays will contain several elements: one or more rgl scenes and controls. Internally rgl uses the combineWidgets function from the manipulateWidget package.

The rgl package provides 3 convenience functions for arranging displays. We have already met the first: the magrittr pipe, %>%. When the display is constructed as a single object using pipes, the objects in the pipeline will be arranged in a single column.

The second convenience function is asRow. This takes as input a list of objects or a combineWidgets object (perhaps the result of a pipe), and rearranges (some of) them into a horizontal row. As in the toggleWidget example, the last argument can be used to limit the actions of asRow to the specified number of components. (If last = 0, all objects are stacked: this can be useful if some of them are not from the rgl package, so piping doesn’t work for them.)

Finally, getWidgetId can be used to extract the HTML element ID from an HTML widget. This is useful when combining widgets that are not all elements of the same pipe, as in the crosstalk example below.

If these convenience functions are not sufficient, you can call manipulateWidget::combineWidgets or other functions from manipulateWidget for more flexibility in the display arrangements.

Integration with crosstalk

The crosstalk package allows widgets to communicate with each other. Currently it supports selection and filtering of observations.

rgl can send, receive and display these messages. An rgl display may have several subscenes, each displaying different datasets. Each object in the scene is potentially a shared dataset in the crosstalk sense.

The linking depends on the rglShared function. Calling rglShared(id), where id is the rgl id value for an object in the current scene, creates a shared data object containing the coordinates of the vertices of the rgl object. This object is passed to rglwidget in the shared argument. It can also be passed to other widgets that accept shared data, linking the two displays.

If a shared data object has been created in some other way, it can be linked to a particular rgl id value by copying its key and group properties as shown in the example below.

# This example requires the crosstalk package.  
# We skip it if crosstalk is not available. 

sd <- SharedData$new(mtcars)
ids <- plot3d(sd$origData(), col = mtcars$cyl, type = "s")
# Copy the key and group from existing shared data
rglsd <- rglShared(ids["data"], key = sd$key(), group = sd$groupName())
rglwidget(shared = rglsd) %>%
asRow("Mouse mode: ", rglMouse(getWidgetId(.)), 
      "Subset: ", filter_checkbox("cylinderselector", 
		                "Cylinders", sd, ~ cyl, inline = TRUE),
      last = 4, colsize = c(1,2,1,2), height = 60)

If multiple objects in the rgl scene need to be considered as shared data, you can pass the results of several rglShared() calls in a list, i.e. rglwidget(shared = <list>). The key values will be assumed to be shared across datasets; if this is not wanted, use a prefix or some other means to make sure they differ between objects.

If the same rgl id is used in more than one rglShared() object, it will respond to messages from all of them. This may lead to undesirable behaviour as one message cancels the previous one.

Low level controls

We repeat the initial plot from this document:

plotids <- with(iris, plot3d(Sepal.Length, Sepal.Width, Petal.Length, 
                  type="s", col=as.numeric(Species)))
subid <- currentSubscene3d()

We might like a button on the web page to cause a change to the display, e.g. a rotation of the plot. First we add buttons, with the “onclick” event set to a function described below:

<button type="button" onclick="rotate(10)">Forward</button>
<button type="button" onclick="rotate(-10)">Backward</button>

which produces these buttons:

We stored the subscene number that is currently active in subid in the code chunk above, and use it as `r subid` in the script below. knitr substitutes the value when it processes the document.

The rotate() function uses the Javascript function document.getElementById to retrieve the <div> component of the web page containing the scene. It will have a component named rglinstance which contains information about the scene that we can modify:

<script type="text/javascript">
var rotate = function(angle) {
  var rgl = document.getElementById("plot3drgl2").rglinstance;
  rgl.getObj(`r subid`).par3d.userMatrix.rotate(angle, 0,1,0);

If we had used webGL=TRUE in the chunk header, the knitr WebGL support would create a global object with a name of the form <chunkname>rgl. For example, if the code chunk was named plot3d2, the object would be called plot3d2rgl, and this code would work:

<script type="text/javascript">
var rotate = function(angle) {
  plot3d2rgl.getObj(`r subid`).par3d.userMatrix.rotate(angle, 0,1,0);


The following functions are described in this document:

ageControl   getWidgetId   propertyControl   subsetControl  
asRow   par3dinterpControl   rglMouse   toggleWidget  
clipplaneControl   playwidget   rglShared   vertexControl