* chore: sync tinygrad
Runs great in sim. now we need to rebuild some models
* oops forgot to unblock this after testing
* helpers
* oh yeah
* latest tg
* this wont do anything empriically
* reduce complexity
* okay lint
* Update tinygrad_runner.py
* Update modeld.py
* Update build-all-tinygrad-models.yaml
* tinygrad bump
* Update modeld.py
* Update tinygrad_runner.py
* bump
* Update SConscript
* Update SConscript
* com
* Update fetcher.py
* Update helpers.py
* life is froughtless, when you're thoughtless
* lint
* ozdust ballroom
* shiz
* Update tinygrad_runner.py
* Update tinygrad_runner.py
* slough it off as i do
* try old support one last time
* support mixed input dtypes
* use internal
* dont need that shiz
* Update fill_model_msg.py
* Update onnx_runner.py
* Update onnx_runner.py
* Update model_runner.py
* see if this speeds up execution
if not, revert me
* no
* Update helpers.py
* rebase
* more
* planplus
---------
Co-authored-by: Jason Wen <haibin.wen3@gmail.com>
* flasher for Rivian long upgrade
* self-contained no dependency on Panda.F4_DEVICES
* standalone flasher
* move to sp module
* use brand field directly
* use brand field directly
* use brand field directly
* bump
* add some logging
---------
Co-authored-by: Jason Wen <haibin.wen3@gmail.com>
* soundd: trigger timeout warning during MADS lateral-only
The selfdrive timeout alert (warningImmediate) only fires when
selfdriveState.enabled is True. During MADS lateral-only mode,
enabled is False even though the system is actively steering.
If selfdrived stops publishing while MADS lateral is active, the
driver gets no audible warning that steering has become unresponsive.
Add selfdriveStateSP to the SubMaster and check mads.active alongside
selfdriveState.enabled so the timeout alert fires whenever the system
is actuating steering.
* test_soundd: add MADS lateral-only timeout test
Test that the selfdrive timeout warning fires when selfdriveState.enabled
is False but selfdriveStateSP.mads.active is True.
---------
Co-authored-by: Jason Wen <haibin.wen3@gmail.com>
The `steer_limited_by_safety` update in `publish()` is gated by
`selfdriveState.active`, which is False during MADS lateral-only
control. This causes the flag to never update once cruise deactivates
— it stays stuck at whatever value it had during the last ramp-up
(typically True), permanently suppressing the saturation timer in
`_check_saturation` and preventing the "Turn Exceeds Steering Limit"
alert from firing.
Use `CC.latActive` instead, which already accounts for MADS via
`get_lat_active()` in ControlsExt.
Bug was introduced in #446 (MADS), which updated `CC.latActive` to
use `mads.active` but missed updating the `steer_limited_by_safety`
gate in `publish()`.
* should fail
* this works but i think i know a better way
* something like this
* hmm
* this works
* rm useless test
* good stash
* Revert "good stash"
This reverts commit c2dddf0810286cb56e2418dd6f7085c2239e5109.
* replace python3-dev apt install with vendored package
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* keep for agnos
* cleaner
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Reapply "feat(lpa): `at` client + list profiles (#37271)" (#37322)
This reverts commit ddf8abc14a.
* lpa: fall back to ModemManager D-Bus when serial port unavailable
On older devices, ModemManager still claims /dev/ttyUSB2, so the
direct serial open fails. Try serial first; if it can't be acquired,
transparently route AT commands through MM's D-Bus Command() interface.
Co-authored-by: Cursor <cursoragent@cursor.com>
* lpa: add serial/dbus transport labels to debug logs
Co-authored-by: Cursor <cursoragent@cursor.com>
* no
* lint
* here
* const
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
github-actions[bot]