diff --git a/R/PipeOpFeatureUnion.R b/R/PipeOpFeatureUnion.R
index ecc05448..cd7dc074 100644
--- a/R/PipeOpFeatureUnion.R
+++ b/R/PipeOpFeatureUnion.R
@@ -57,6 +57,26 @@ cbind_tasks = function(inputs, assert_targets_equal) {
     stopf("All tasks must have the same target columns")
   }
 
+  if (is.null(names(inputs))) names(inputs) = paste0("input", seq_along(inputs))
+  input_names = map(inputs, function(input) input$feature_names)
+  if (any(duplicated(unlist(input_names)))) {
+    # Treat duplicate names: Suffix with name of the input
+    new_names = imap(input_names, function(input, name) {
+      dupe = input %in% unlist(input_names[!(names(input_names) == name)])
+      input[dupe] = paste(input[dupe], name, sep = "_")
+      if (any(dupe)) return(input) else return(NULL)
+    })
+    new_names = discard(new_names, is.null)
+    # Use replace_features on renamed data. This seems very inefficient
+    inputs[names(new_names)] = imap(inputs[names(new_names)], function(input, name) {
+      browser()
+      dt = input$data(ids, input$feature_names)
+      colnames(dt) = new_names[[name]]
+      input$replace_features(dt)
+      return(input)
+    })
+  }
+
   new_cols = Reduce(function(x, y) rcbind(x, y$data(ids, y$feature_names)), tail(inputs, -1L), init = data.table())
   task$clone(deep = TRUE)$cbind(new_cols)
 }
diff --git a/tests/testthat/test_pipeop_featureunion.R b/tests/testthat/test_pipeop_featureunion.R
index 36de060d..f90c613e 100644
--- a/tests/testthat/test_pipeop_featureunion.R
+++ b/tests/testthat/test_pipeop_featureunion.R
@@ -75,6 +75,21 @@ test_that("Test wrong inputs", {
   task = mlr_tasks$get("iris")
   expect_error(g$train(task), "Assertion on 'rows'")
 })
+
+# FIXME: Somewhat depends on https://github.com/mlr-org/mlr3/issues/268
+# test_that("Duplicate Features", {
+#   # Define PipeOp's
+#   tsk = mlr_tasks$get("iris")
+#   t1 = tsk$clone()$set_col_role("Sepal.Length", character())
+#   t2 = tsk$clone()$set_col_role(c("Petal.Length", "Petal.Width"), character())
+
+#   po = PipeOpFeatureUnion$new(2)
+
+#   tout = train_pipeop(po, list(t1, t2))
+#   expect_equivalent(tout[[1]]$feature_names, c())
+#   expect_equivalent(tout[[1]]$target_names, tsk$target_names)
+# })
+
 # FIXME: depends on mlr-org/mlr3#179
 ## test_that("PipeOpFeatureUnion - levels are preserved", {
 ##
diff --git a/vignettes/examples.Rmd b/vignettes/examples.Rmd
index 03036d67..4c071d92 100644
--- a/vignettes/examples.Rmd
+++ b/vignettes/examples.Rmd
@@ -453,7 +453,7 @@ Additionally, we also keep a version of the level 0 output (via `PipeOpNull`) an
   # Plot the resulting graph
   level_2$plot(html = TRUE)
 
-  task = mlr_tasks$get("iris"),
+  task = mlr_tasks$get("iris")
   lrn = GraphLearner$new(level_2)
 
   lrn$