On a Luma P1, autoexposure on first boot was getting overridden with the manual exposure setting. This was traced back to #1814, where the order of setting auto exposure and raw exposure was flipped. This flips it back.
## Description
Back in
https://github.com/PhotonVision/photonvision/pull/467#issuecomment-1280964529
and
https://discord.com/channels/725836368059826228/725846784131203222/974498049609056266
we added code to poke our NT client every 5 seconds to "clicking the
save button in the settings window makes Photon show up again over
networktables". Total hack, but it seemed to work. We didn't at the time
dig any deeper in Wireshark or debug-level NT logs.
<img width="1373" height="679" alt="image"
src="https://github.com/user-attachments/assets/8cb2102e-0bae-4bfd-b9ac-55d31f8421b6"
/>
Now, it's 2026. 4 years on from the OG bug. And this code seems linked
to these issues
-
https://www.chiefdelphi.com/t/photonvision-coprocessor-not-sending-data-can-t-change-networking/516356/5
-
https://www.chiefdelphi.com/t/photonvision-network-tables-known-issue/515966
Craig collected these log files as well:
[craig-nt-never-connects.zip](https://github.com/user-attachments/files/26001809/craig-nt-never-connects.zip)
The code path that handles TCP re-connection was also changed entirely
since we first added this workaround. Regardless this hack was not
removed as part of the NT3 to NT4 upgrade:
- pre-NT4, reconnection was handled by `TCPConnector::connect_parallel`
which delegates to `TCPConnector`. This uses raw socket APIs
- post-NT4, reconnect is handled by `ParallelTcpConnector`. This uses
libuv exclusively
@crschardt did some really great debugging with a rio and radio in the
loop with a luma p1. In this test setup, removing this code improves
stability markedly. I'd still like to get this more on robot time, as
well as try to understand from Peter why we might have needed this code
in the first place.
## Changes
- Remove periodic stop/restart of NetworkTables client every 5 seconds
if `NetworkTablesInstance::isConnected` returns false.
## Meta
Merge checklist:
- [ ] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [ ] The description documents the _what_ and _why_, including events
that led to this PR
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with all settings going back to the previous seasons's last release
(seasons end after champs ends)
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
- [ ] If this PR adds a dependency, the license has been checked for
compatibility and steps taken to follow it
---------
Co-authored-by: Craig Schardt <crschardt@fastem.com>
## Description
<!-- What changed? Why? (the code + comments should speak for itself on
the "how") -->
<!-- Fun screenshots or a cool video or something are super helpful as
well. If this touches platform-specific behavior, this is where test
evidence should be collected. -->
<!-- Any issues this pull request closes or pull requests this
supersedes should be linked with `Closes #issuenumber`. -->
Pursuant to PhotonVision/rubik_jni#21 modify the neuralnetworkresult and related code to use a rotatedrect. This is scaffolding for implementing OBB -- overall this shouldn't change behavior of existing normal object detection models. This PR also bumps the rubik_jni version.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
Co-authored-by: Matt Morley <matthew.morley.ca@gmail.com>
Given that WPILib is nuking their Java tooling, it becomes necessary to
pull said tooling into PV itself. This migrates the
CombinedRuntimeLoader into PV, which should finalize all of the tooling
migration.
## Description
Presently, any tests using our old configs happen in place. This is
problematic, as it changes the files themselves. This means that anyone
running the tests will cause unintentional modifications, which stand a
chance of being committed and merged into main. We want these old
database files to remain untouched, thus we copy them to a temp
directory prior to running our tests.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_, including events
that led to this PR
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with all settings going back to the previous seasons's last release
(seasons end after champs ends)
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
## Description
Previously, when attempting to load a model that did not follow the old
naming pattern that was in the models directory, we caused PV to crash
due to a missing early return (reproducable on this branch with the
extra bonus ML model and missing label file):
```
[2026-02-01 17:35:15] [Config - NeuralNetworkModelManager] [ERROR] Model properties are null. This could mean the config for model /home/matth/photonvision/test-resources/old_configs/2025.3.1-old-nnmm/models/iCauseProblems.rknn was unable to be found in the database. Trying legacy...
[2026-02-01 17:35:15] [Config - NeuralNetworkModelManager] [ERROR] Failed to translate legacy model filename to properties: /home/matth/photonvision/test-resources/old_configs/2025.3.1-old-nnmm/models/iCauseProblems.rknn: /home/matth/photonvision/test-resources/old_configs/2025.3.1-old-nnmm/models/iCauseProblems-labels.txt
SQLConfigTest > testLoadNewNNMM() FAILED
java.lang.NullPointerException: Cannot invoke "org.photonvision.common.configuration.NeuralNetworkModelsSettings$ModelProperties.toString()" because "properties" is null
at org.photonvision.common.configuration.NeuralNetworkModelManager.loadModel(NeuralNetworkModelManager.java:342)
at java.base/java.util.stream.ForEachOps$ForEachOp$OfRef.accept(ForEachOps.java:183)
```
This PR fixes it by adding the missing early-return, and adding unit
tests to make sure we handle this.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_, including events
that led to this PR
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with all settings going back to the previous seasons's last release
(seasons end after champs ends)
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [x] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: Matt M <matthew.morley.ca@gmail.com>
## Description
#2255 introduced a new, cross-platform method for monitoring hardware
and removed the custom shell commands that had been used previously.
This PR updates the documentation to reflect the removal of those
commands from hardwareConfig.json.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_, including events
that led to this PR
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with all settings going back to the previous seasons's last release
(seasons end after champs ends)
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: Gold856 <117957790+Gold856@users.noreply.github.com>
## Description
In https://github.com/PhotonVision/photonvision/issues/1771, we
discovered that the OV9281 on Linux seems to sometimes ignore our
exposure requests on first boot if we're in manual mode. Cycling the
camera's auto exposure with pauses in-between seems to fix it, which is
what this PR does.
Tested by [Team 2881 on
Discord](https://discord.com/channels/725836368059826228/725846784131203222/1465098110765105456);
thanks for helping us test this fix out!
## Meta
Closes https://github.com/PhotonVision/photonvision/issues/1771
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2024.3.1
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: Gold856 <117957790+Gold856@users.noreply.github.com>
## Description
<!-- What changed? Why? (the code + comments should speak for itself on
the "how") -->
<!-- Fun screenshots or a cool video or something are super helpful as
well. If this touches platform-specific behavior, this is where test
evidence should be collected. -->
<!-- Any issues this pull request closes or pull requests this
supersedes should be linked with `Closes #issuenumber`. -->
closes#2318closes#2319
For the 2027 game, teams might want to detect more than 10 results.
Therefore, we're increasing the limit.
We also ran into an issue with our sim, where a user can create too many
objects and cause an overflow. We implement that same limit of 50
targets here.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
## Description
Monitoring CPU Temperature on Windows is challenging because most
vendors don't publish this data to WMI. As a work-around, OSHI tries to
use
[LibreHardwareMonitor](https://github.com/LibreHardwareMonitor/LibreHardwareMonitor)
via
[jLibreHardwareMonitor](https://github.com/pandalxb/jLibreHardwareMonitor).
If the temperature isn't found in WMI and jLibreHardwareMonitor isn't
present, OSHI issues warnings every time `getCpuTemperature()` is
called. This clogs the console with useless information when running on
Windows and makes testing difficult.
We could include jLibreHardwareMonitor as a dependency for our Windows
jar, but LibreHardwareMonitor installs Winring0.sys, which is a kernel
level driver with an unfixed severe vulnerability. Windows defender
flags Winring0 as a vulnuratble driver and blocks it from installing.
Rather than messing with it, this PR prevents Windows systems from
calling the `getCpuTemperature()` method in OSHI.
Fixes#2280
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
---------
Co-authored-by: Gold856 <117957790+Gold856@users.noreply.github.com>
## Description
Update to use our fork of the WPILib tool plugin. WPILib has indicated
that it won't be maintained past 2026, so we're forking it under our
org. We also fixed a bug that results in native libraries for multiple
platforms being included in a jar when only the native libraries for the
platform being built should be included.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: Gold856 <117957790+Gold856@users.noreply.github.com>
Co-authored-by: samfreund <samf.236@proton.me>
## Description
[OSHI](https://github.com/oshi/oshi) is a free (MIT license) JNA-based
library for accessing hardware and system performance information. This
PR includes a re-write of the metrics monitoring code to be based on
OSHI. The original intent was to gain access to data about network
traffic for addition to the Settings tab. An additional benefit is that
collecting the data is now around two orders of magnitude (or more)
faster!
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [x] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [x] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: samfreund <samf.236@proton.me>
## Description
We currently log some things with ERROR status and include stack traces
for events that are typical behavior. This pollutes the logs and makes
it harder to track down real errors.
This PR changes the way that some events are logged:
* missing configs in the database are logged as [INFO] without the
exception stack trace.
* skip parsing NPU usage when the command is blank so that it doesn't
throw a NumberFormatException.
* log warn instead of error for unsupported NN backends (added by
@samfreund)
* skip warn when we don't add a model, only debug when we add it (added
by @samfreund)
Before:
```
Oct 22 20:56:26 photonvision java[831]: [2024-10-22 20:56:26] [Config - SqlConfigProvider] [ERROR] Could not deserialize apriltag layout! Loading defaults: Provided empty string for class edu.wpi.first.apriltag.AprilTagFieldLayout
Oct 22 20:56:26 photonvision java[831]: [2024-10-22 20:56:26] [Config - SqlConfigProvider] [ERROR] org.eclipse.jetty.io.EofException: Provided empty string for class edu.wpi.first.apriltag.AprilTagFieldLayout
Oct 22 20:56:26 photonvision java[831]: at org.photonvision.common.util.file.JacksonUtils.deserialize(JacksonUtils.java:136)
Oct 22 20:56:26 photonvision java[831]: at org.photonvision.common.configuration.SqlConfigProvider.load(SqlConfigProvider.java:298)
Oct 22 20:56:26 photonvision java[831]: at org.photonvision.common.configuration.ConfigManager.load(ConfigManager.java:198)
Oct 22 20:56:26 photonvision java[831]: at org.photonvision.Main.main(Main.java:290)
```
After:
```
Dec 15 22:29:09 photonvision java[662]: [2025-12-15 22:29:09] [Config - SqlConfigProvider] [INFO] Missing AprilTag Field Layout in database. Loading defaults
```
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: samfreund <samf.236@proton.me>
Co-authored-by: Gold856 <117957790+Gold856@users.noreply.github.com>
## Description
On the RubikPi3, PhotonVision starts before NetworkManager has fully
initialized. This causes it to fail to identify the network interfaces
available on the board, which leads to problems with configuring and
controling networking from the UI. The failure can be detected by the
call to `nmcli` returning an [exit status of
8](https://networkmanager.dev/docs/api/latest/nmcli.html#exit_status),
which means "NetworkManager is not running."
This PR retries the call to nmcli every 0.5 seconds until the exit
status does not equal 8, or a maximum of 10 attempts have been made. The
retry only occurs the first time `getAllInterfaces()` is called.
Subsequent calls to this method will only make one attempt to avoid
locking up the program if networking isn't responding as expected.
In my testing on the RubikPi3, the code has to retry for less than 2
seconds in order to get a valid response from NetworkManager.
The need for this is greatly reduced by
https://github.com/PhotonVision/photon-image-modifier/pull/114, but this
code adds an additional layer of robustness against slow network
startup.
Closes#2212
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: samfreund <samf.236@proton.me>
## Description
On boot, the interface might not be fully configured yet, and the
returned NetworkInterface will be null. This would cause an NPE to be
thrown when attempting to get the MAC address from the interface. This
checks if the interface is null and silently returns an empty string if
it is, which is okay if the networking is being managed because if that
interface is broken, there's bigger problems. Fixes#2244.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
I have all sorts of weird adapters from npcap and Hyper-V that don't
have MAC addresses, so I'm simplifying the logic down so that it always
tries to find _any_ adapter with a MAC address, but attempts to see if
it can find what adapter is in use right now and use the MAC address
from that. This also unpublishes old MAC address topics, which wasn't
done before.
The old GPIO abstraction was replaced with
[`diozero`](https://www.diozero.com), which supports most hardware
running Linux due to its use of GPIO character devices provided by the
Linux kernel. `diozero` also supports [alternate
providers](https://www.diozero.com/concepts/providers.html#providers) if
for some reason the character device API is insufficient. Certain
capabilities outside of the character device API is also implemented for
common hardware.
Custom GPIO commands are implemented via a custom `diozero` provider.
The configuration for custom GPIO will need manually updated according
to the Hardware Config documentation page.
The status LED class was also reworked to support additional statuses or
LED indication types, although none have been added yet.
This was tested on a RPi 5 with LL3 illumination LEDs and an RGB status
led attached. All capabilities worked as expected. All 8 status LED
colors were tested and functional via modifying the code. Basic
functionality of custom GPIO was tested with dummy commands.
## Description
#2224 removed the custom deserializers for `Path`, but we still need one
to be able to deserialize the `Path` key in
`NeuralNetworkPropertyManager`. Additionally, Jackson seems to
auto-convert the `Path` key to a `String` using `toString` instead of
its own serializers, so a custom key serializer is also needed to
consistently use the same format for paths. This also removes unused
serde methods in `JacksonUtils` to minimize potential future churn, and
tacks an `@JsonIgnore` on `getModels` to prevent Jackson from
serializing a `ModelProperty` array into the database.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [x] If this PR addresses a bug, a regression test for it is added
## Description
PhotonJNICommon is just our implementation of combined runtime loader,
which we don't really need. This removes it and just uses
CombinedRuntimeLoader directly. This also fixes the issues introduced in
#2219, which lead to some of our JNIs not loading.
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [x] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: Matt M <matthew.morley.ca@gmail.com>
## Description
We can avoid copying files by chunks just using `Files.copy`. This
should be NFC, just makes the code cleaner
<!-- What changed? Why? (the code + comments should speak for itself on
the "how") -->
<!-- Fun screenshots or a cool video or something are super helpful as
well. If this touches platform-specific behavior, this is where test
evidence should be collected. -->
<!-- Any issues this pull request closes or pull requests this
supersedes should be linked with `Closes #issuenumber`. -->
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [ ] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
## Description
Updating to take advantage of the now independent mrcal-java (no longer
need to install SuiteSparse and friends!)
## Meta
Merge checklist:
- [x] Pull Request title is [short, imperative
summary](https://cbea.ms/git-commit/) of proposed changes
- [x] The description documents the _what_ and _why_
- [ ] If this PR changes behavior or adds a feature, user documentation
is updated
- [ ] If this PR touches photon-serde, all messages have been
regenerated and hashes have not changed unexpectedly
- [ ] If this PR touches configuration, this is backwards compatible
with settings back to v2025.3.2
- [ ] If this PR touches pipeline settings or anything related to data
exchange, the frontend typing is updated
- [ ] If this PR addresses a bug, a regression test for it is added
---------
Co-authored-by: Matt Morley <matthew.morley.ca@gmail.com>
