Phase out NUgus and introduce Booster K1 to the codebase by ZeoNyph · Pull Request #1828 · NUbots/NUbots

ZeoNyph · 2026-09-07T06:39:44Z

We all knew this day was coming! This PR is the culmination of months of work and phases out the NUgus for the K1.

There are a lot of changes:

Compilation toolchain: This is one of the bigger changes, and took me the longest time to actually get ready, as the compilation toolchain was reworked to allow for cross-compilation across processor architectures (x86 to ARM). I also had to change the base to Ubuntu 22.04 (goodbye Arch my beloved ;-;) to match the distro used by the K1s, and remove/add dependencies where needed. Have a look at the stuff in /docker to see the shennanigans I had to get into.
Booster SDK integration: I had to add the Booster SDK to the Docker image (and update the version constantly as Booster loves to add new and useful features every major update lol), and also write a new Booster HardwareIO module so that we can integrate it with the higher-level modules. I tried to make it so that the Booster module would be the only module that directly uses the SDK, but alas, K1Camera and K1Sensor failed to adhere to that. Speaking of...
New K1-specific modules: there are now K1-specific modules for the camera and sensor data, as the NUgus RawSensors module had a lot of bespoke logic that was made with the NUgus in mind, and the camera relies on DDS messaging as opposed to the previous USB stream with the Aravis library. @Tom0Brien helped out a lot here when it came to rewriting the sensors module, so thank you Tom!
TensorRT as main vision inference library: Given that we are moving to a platform that runs on an NVIDIA Jetson Orin NX, OpenVINO can no longer cut it as the main execution provider for ONNX inference (given that it doesn't work at all on the Jetson). As a result, I had to make a TensorRT execution provider for YOLO so we could get it running on the Jetson.
Behaviour changes: a lot happened here, primarily as a result of the collaboration with UNSW and RMIT, and so @sebastianbrooker2 refactored the Support module to rely on a newly-made formation.yaml file made from the formation rules used by the other teams. There were changes here that I'm not fully across, so let me know if I missed anything here.
Visual Mesh disabled: unfortunately, for now we've had to disable Visual Mesh, as OpenCL does not work on the Jetson, and CPU inference was too slow for gameplay (1 fps was what we saw).
SystemConfiguration and RoboCupConfiguration changes: these had to be changed in order to account for the fact that we do not setup the K1s from scratch (meaning we don't have a fresh OS install and install scripts like we do for the NUgus). This primarily applies to SystemConfiguration, but RobocupConfiguration has changes made as part of the work done at Incheon and Beijing (NetworkManager instead of systemd-network, ability to restart Booster's daemon, ability to use DHCP, and more)

There were a lot of changes done, and I don't fully recall all of them, so I'll add them in as I remember.

Reviewers, Note

I will need to do a full pass of NUbook documentation and make a PR for that as well (including archiving NUgus and Darwin documentation). There's also a fair bit of work left on the K1 repo in terms of outstanding PRs and changes (need to bring in changes from Beijing into main, and also look into things such as switching to ONNXRuntime instead of maintaining a bespoke TensorRT implementation).

There's also ongoing work with the NUgus (@vincenttumminello is working on RL with the NUgus as part of his FYP), as well as other outstanding PRs on this end of the codebase.

Once both ends have merged/closed their relevant PRs, we can tag the most recent NUgus commit and begin the merge.

Things left to do

Write NUbook documentation
Cleanup K1 repo (outstanding PRs)
Ensure all NUgus work is complete
Other miscellaneous items I have missed

…. tidy proto

…ility

Tom0Brien · 2026-09-07T07:02:50Z

Tom0Brien · 2026-09-07T07:03:01Z

Tom0Brien · 2026-09-07T07:04:48Z

remove "K1" from the start of the skill names

Tom0Brien · 2026-09-07T07:06:08Z

don't thing this should be here

Tom0Brien · 2026-09-07T07:06:28Z

don't think this should be here

Tom0Brien · 2026-09-07T07:08:05Z

can probably remove tinyrobotics as a dep now I think

Tom0Brien · 2026-09-07T07:09:24Z

can be renamed to just Camera

Tom0Brien · 2026-09-07T07:11:05Z

Should try get CI working

ysims

Partial review.

ysims · 2026-09-07T07:04:22Z

    strategy:
      matrix:
-        platform: ["generic", "nuc12wshi7"]
+        platform: ["generic_k1", "orinnx"]


Suggested change

platform: ["generic_k1", "orinnx"]

platform: ["generic", "orinnx"]

this could still be called generic, right?

yes it can; I changed it so that if I ever got Actions running on the K1 repo it wouldn't interfere with the NUgus generic image, but it can be changed back for the merge in

ysims · 2026-09-07T07:04:25Z

+          build-args: |
+            platform=generic


ysims · 2026-09-07T07:04:42Z

+          build-args: |
+            platform=${{ matrix.platform == 'generic_k1' && 'generic' || matrix.platform }}


new toolchain uses the platform argument to determine whether to setup stuff for cross-compilation (ARM sysroot, apt repository mirrorlist, etc.), hence the build arg

ysims · 2026-09-07T07:04:54Z

+          build-args: |
+            platform=generic


ysims · 2026-09-07T07:05:34Z

 include(ToolchainLibraryFinder)
 ToolchainLibraryFinder(
-  NAME zlib
+  NAME ZLIB


why's the name caps now

zlib wouldn't work if it wasn't all caps for some reason; will need to double check

ysims · 2026-09-07T07:21:17Z

+    // Must match the writer-side layout in NUbridge exactly (binary compatibility).
+    struct SharedImageHeader {


Pull all of this out into a folder similar to what is done for the sensor filter - try to just have the Reactor constructor in here.

K1Camera got a refactor in the beijing-masters-2026 branch to no longer need a shared memory ROS2 bridge; will need to get that in ASAP since it changes a lot of the code you commented on

ysims · 2026-09-07T07:23:52Z

+        });
+
+        on<Trigger<Sensors>>().then("Buffer Hcw", [this](const Sensors& sensors) {
+            std::lock_guard<std::mutex> lock(sensors_mutex);


Is this trying to stop this reaction running more than once at a time? Use Single instead, or look at Sync.

ysims · 2026-09-07T07:24:02Z

+            std::lock_guard<std::mutex> lock(sensors_mutex);
+            auto now = NUClear::clock::now();
+            Hcws.resize(std::distance(Hcws.begin(), std::remove_if(Hcws.begin(), Hcws.end(), [now](const auto& v) {
+                                          return v.first < (now - std::chrono::milliseconds(500));


magic number?

ysims · 2026-09-07T07:26:30Z

+        std::mutex cameras_mutex;
+        std::mutex sensors_mutex;
+        std::vector<std::unique_ptr<CameraContext>> cameras;
+        std::vector<std::pair<NUClear::clock::time_point, Eigen::Isometry3d>> Hcws;
+
+        void camera_thread(CameraContext& ctx);
+        void stop_cameras();


documentation directives

ysims · 2026-09-07T07:27:09Z

+head_pose:
+  - segment: _head_pose
+
+# Deadband (metres for x/y, radians for theta) below which normalized odometry components are


Suggested change

# Deadband (metres for x/y, radians for theta) below which normalized odometry components are

# Deadband (metres for x/y, radians for theta) below which normalised odometry components are

Tom0Brien and others added 30 commits June 22, 2026 21:22

Wrap tiny robotics stuff to solve llvm memory issue

bbbaabe

Update keyboardwalk.role

fff6ecb

Update FieldLocalisationNLopt.yaml

2c5ffce

Update FieldLocalisationNLopt.yaml

8d594c4

Merge branch 'main' into sivanesan/fix-configuration

b325777

Tidy up and remove tinyrobotics dep

ad3d41c

change up SystemConfiguration to work with NetworkManager and Booster

1651950

Update README

3ac3eb5

added nmconfigs for second robot

35e7c30

add second robot 'name'

7dc9a32

Update FieldLocalisationNLopt.yaml

3ab91d3

Update robot sensor pipeline to work with booster sdk

a9dbaa1

Merge branch 'main' into brooker/visualkick

a988775

Fix K1 Visual Kick (#12)

17cc76a

Add temp robot names

29a70a1

add functionality to replace robot name in Booster app

177e6cb

Updated parts of RobocupConfiguration to work with K1

9f46c19

Update SystemConfiguration and RobotConfiguration to work with K1 (#16)

239ce2b

Merge branch 'NUbots:main' into main

2aee548

add shared_message.proto

3a646e1

obvious changes to swap robocup message for shared team message

cd3d69b

downstream things that need name change, removed things that use cost…

daae398

…. tidy proto

remove this no duplicate package and can test build

1e2cc06

fixing things so it builds

7cf588b

Refactor Dockerfile and platform handling for generic target compatib…

4042e55

…ility

add formation.yaml to support purpose and change hpp as needed

5ee6b6e

change support logic to follow formation rules

c358306

whoops

c4ddc72

claude helping me fix things

b361ba1

priority of attack based on ID

fd91a4a

ZeoNyph requested review from jpptm, sebastianbrooker2, vincenttumminello and yippyy September 7, 2026 06:39

ZeoNyph self-assigned this Sep 7, 2026

Tom0Brien reviewed Sep 7, 2026

View reviewed changes

Comment thread module/vision/VisualMesh/data/config/VisualMesh.yaml

Tom0Brien Sep 7, 2026

Copy link
Copy Markdown

Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

remove

ZeoNyph reacted with thumbs up emoji

Tom0Brien reviewed Sep 7, 2026

View reviewed changes

Tom0Brien requested changes Sep 7, 2026

View reviewed changes

ysims reviewed Sep 7, 2026

View reviewed changes

Beijing Masters 2026 changes (#53)

ea6e3a3

	platform: ["generic_k1", "orinnx"]
	platform: ["generic", "orinnx"]

		build-args: \|
		platform=${{ matrix.platform == 'generic_k1' && 'generic' \|\| matrix.platform }}

		// Must match the writer-side layout in NUbridge exactly (binary compatibility).
		struct SharedImageHeader {

	# Deadband (metres for x/y, radians for theta) below which normalized odometry components are
	# Deadband (metres for x/y, radians for theta) below which normalised odometry components are

Uh oh!

Conversation

ZeoNyph commented Sep 7, 2026

Reviewers, Note

Things left to do

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Tom0Brien commented Sep 7, 2026

Uh oh!

ysims left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ZeoNyph Sep 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ZeoNyph Sep 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

ZeoNyph Sep 7, 2026 •

edited

Loading

ZeoNyph Sep 7, 2026 •

edited

Loading