Commit Graph

84 Commits

Author SHA1 Message Date
ZwX1616
ba2d2677c1 modeld: no hardcoded frame names (#35476)
* from model

* juggle
2025-06-06 13:05:54 -07:00
Kacper Rączy
97ad175243 Reapply "Enable online lag learning (#34968)" (#35327)
* Revert "Revert "Enable online lag learning (#34968)" (#35322)"

This reverts commit 9c40c48729.

* Update ref commit

* Update ref

* Update ref commit
2025-05-24 02:31:17 +00:00
Kacper Rączy
9c40c48729 Revert "Enable online lag learning (#34968)" (#35322)
* Revert "Enable online lag learning (#34968)"

* Update ref commit
2025-05-22 01:55:19 +00:00
Adeeb Shihadeh
350a235303 modeld: more USB GPU fixes (#35306)
* fixups

* builds
2025-05-20 19:41:58 -07:00
Adeeb Shihadeh
d0bf2be6f0 External GPU support for big models (#35172)
* usb gpu

* cleanup

---------

Co-authored-by: Comma Device <device@comma.ai>
2025-05-13 17:12:32 -07:00
Adeeb Shihadeh
67486ff92d bump tinygrad (#35208)
* bump tinygrad

* fix

* why is mac different?

* fix sim

* relax that
2025-05-13 16:59:35 -07:00
Harald Schäfer
e9cea3ae5c Revert TR again (#35179)
Revert "Tomb Raider 7 (#35114)"

This reverts commit 0fb4aafa35.
2025-05-11 08:24:46 -07:00
Harald Schäfer
0fb4aafa35 Tomb Raider 7 (#35114)
* Revert "Revert TR (#35110)"

This reverts commit df4f2955dc.

* eb5f884a-10ad-49fd-ae5c-e2818c26e568/400

* 1cc828ab-95e5-4620-aa07-b98918b4268d/400

* 5790a2c1-b487-4bef-a3c3-db1fcd5a756d/400

* raw plan
2025-05-07 18:06:41 -07:00
Harald Schäfer
df4f2955dc Revert TR (#35110)
* Revert "Tomb raider 2 (#35029)"

This reverts commit 2c162d9b75.

* bugfix

* fix policy

* min control speed
2025-05-02 20:54:42 -07:00
Harald Schäfer
e1d2360b8c Revert: Tomb Raider 6 (#35105) 2025-05-02 10:17:40 -07:00
Kacper Rączy
45b273328d Enable online lag learning (#34968)
* Use in modeld

* Use in torqued

* Do not use default value in torqued

* Add liveDelay in process replay

* update ref

---------

Co-authored-by: Bruce Wayne <harald.the.engineer@gmail.com>
2025-05-01 11:42:49 -07:00
Harald Schäfer
9f1e462faa Tomb Raider 6 (#35087)
* 5ec366c3-7883-4004-84a2-e4b14bac5b1d/400

* Use lat plan

* fix import

* fix

* 8d0a1b3b-9972-4e53-b9c5-3e13e5e3e404/400

* whitespace

* whitespace
2025-04-30 17:19:53 -07:00
Harald Schäfer
756aebd39f Tomb Raider 4 (#35058)
* dc455ed7-0ef1-4c57-9391-c71767784c7e/400

* f16884e6-417e-4842-aeda-b2e1b3edf40b/400
2025-04-24 16:24:21 -07:00
Harald Schäfer
a38dcbb3fe Disable low speed steer in modeld (#35056)
* disable low speed steer

* No steer under
2025-04-23 14:15:06 -07:00
Harald Schäfer
8cefc00a6e Tomb Raider 3 (#35042)
* 92b64884-4506-4a03-87ad-33e1a177fe73/400

* faster lat
2025-04-21 17:06:53 -07:00
Harald Schäfer
33a7d853f0 Modeld: small refactor (#35044)
refactor
2025-04-21 15:10:41 -07:00
Harald Schäfer
2c162d9b75 Tomb raider 2 (#35029)
* db56b8fb-9135-4ab6-af18-99b7df7b2245/400

* fixes

* linter unhappy

* 6dbe0991-baa1-49ad-836a-ab370d1f0d92/400

* This one is good: 19387087-1005-475e-9015-9458dd8e7c5f/400

* Better every day: 39ed911c-0937-417f-97d2-58a8bb3caa53/400

* Actually end-to-end

* typo

* smooooooth: 94e23541-eb84-4fef-9f51-6a2d82aff314/360

* Revert "smooooooth: 94e23541-eb84-4fef-9f51-6a2d82aff314/360"

This reverts commit edd4f02386d83d82dd8a188985cde80ed1646b7f.

* 11632ef7-f555-489c-8480-e3bf97d9285e/400

* 08712d27-f6bd-4536-a30e-c729e5f62356/400

* 0a92a35e-1f72-476a-8cb6-c9f103f36822/400

* ee6d2394-2072-420c-a664-b4c0d4ed0b61/400

* no prev curv

* No double work

* fix bug

* smooth

* update prev action

* whitespace

* add little accel

* new ref

* Update plant.py
2025-04-17 23:21:25 -07:00
Harald Schäfer
b42ec33a63 Update modeld action logic (#35032)
* add action logic

* magic numbers

* unused

* plot accel

* more changes
2025-04-17 17:59:15 -07:00
ZwX1616
d807b5c476 Filet o Fish model (#34798)
* 690b01c3 seems ok

* correct temporal

* push

* inplace

* bs

* what thw

* is this wrong

* frames are skipped

* new models

* simplify decimation

* clean up

* clean up modelframe

* need attr

* lint

* 0

* use all samples

* this should break - Revert "use all samples"

This reverts commit 6c0d7943ac5fbb7ae60af1a1201e2423e4c3c105.

* add lc probs

* Revert "this should break - Revert "use all samples""

This reverts commit ca38c54333555266a0d2c885c28af28941841431.

* Reapply "this should break - Revert "use all samples""

This reverts commit a3f0246f209f85f06b9090d9492bfba32ed8cfed.

* Revert "Reapply "this should break - Revert "use all samples"""

This reverts commit 7fd3d2a191b688e5ef7b4dcc8f5379e900af10f8.

* new fish

* e07ce1de-bdea-463e-b5bc-a38ce8d43f4f/700

---------

Co-authored-by: Comma Device <device@comma.ai>
2025-03-07 14:24:22 -08:00
Shane Smiskol
a82db2d331 Revert "Filet o Fish model" (#34780)
Revert "Filet o Fish model (#34775)"

This reverts commit 31076b818d.
2025-03-05 01:41:16 -08:00
ZwX1616
31076b818d Filet o Fish model (#34775)
* 690b01c3 seems ok

* correct temporal

* push

* inplace

* bs

* what thw

* is this wrong

* frames are skipped

* new models

* simplify decimation

* clean up

* clean up modelframe

* need attr

* lint

* 0

* use all samples

* this should break - Revert "use all samples"

This reverts commit 6c0d7943ac5fbb7ae60af1a1201e2423e4c3c105.

* add lc probs

* Revert "this should break - Revert "use all samples""

This reverts commit ca38c54333555266a0d2c885c28af28941841431.

* Reapply "this should break - Revert "use all samples""

This reverts commit a3f0246f209f85f06b9090d9492bfba32ed8cfed.

* Revert "Reapply "this should break - Revert "use all samples"""

This reverts commit 7fd3d2a191b688e5ef7b4dcc8f5379e900af10f8.

---------

Co-authored-by: Comma Device <device@comma.ai>
2025-03-04 20:35:57 -08:00
Adeeb Shihadeh
e5ba737db0 Revert "Filet o Fish model (#34637)"
This reverts commit 159b1c9eb4.
2025-03-03 22:56:59 -08:00
ZwX1616
159b1c9eb4 Filet o Fish model (#34637)
* 690b01c3 seems ok

* correct temporal

* push

* inplace

* bs

* what thw

* is this wrong

* frames are skipped

* new models

* simplify decimation

* clean up

* clean up modelframe

* need attr

* lint

* 0

---------

Co-authored-by: Comma Device <device@comma.ai>
2025-03-03 15:15:36 -08:00
ZwX1616
29ca4aa7de modeld refactor: supercombo -> vision+policy (#34718)
* should be ok

* update tg

* type

* wrong model

* is no more

* update readme

* add back

---------

Co-authored-by: Comma Device <device@comma.ai>
2025-02-27 17:00:56 -08:00
Harald Schäfer
d41311ca20 Switch PC running to tinygrad with LLVM (#34660)
* try llvm on cpu

* update lock

* Make sure test fails

* With BEAM=3 its fast

* small fixes

* more fixes

* Fix macos

* do dm too

* rm ort
2025-02-24 14:51:10 -08:00
Shane Smiskol
1d86afdf83 carName -> brand (#34474)
* bump

* brand

* body is lower
2025-01-25 01:39:06 -06:00
Harald Schäfer
c8264cbf2c Notre Dame model in tinygrad (#34324)
* release model: 6f23a03f-486b-4d3e-a314-19d149644c7c/700

* old style model in tinygrad

* fix desire

* tg hack

* 20Hz

* no gas probs

* No gas here

* better indexing

---------

Co-authored-by: Yassine Yousfi <yyousfi1@binghamton.edu>
2025-01-06 20:52:36 -08:00
Harald Schäfer
17ca6389e1 Tinygrad runner (#34261)
* squash

* dmonitoringmodeld: use cl transform (#34235)

* needs cleanup

* only if tici

* bump tinygrad

* check width

* base modelframe

* .

* need to be args

* more cleanup

* no _frame in base

* tici only

* its DrivingModelFrame

* .6 is fair

---------

Co-authored-by: Comma Device <device@comma.ai>

* Update tinygrad

* tg upstream

* bump tg

* bump tg

* debug

* attr

* misc cleanup

* whitespace

* remove

* Add TODOs to make python proc for modelrunners

* whitespace

---------

Co-authored-by: ZwX1616 <zwx1616@gmail.com>
Co-authored-by: Comma Device <device@comma.ai>
Co-authored-by: Maxime Desroches <desroches.maxime@gmail.com>
2024-12-18 11:58:59 -08:00
ZwX1616
7352e612a2 dmonitoringmodeld: use cl transform without tinygrad (#34266)
* merge

* why

* self.buf_size

* 0.05 more than with tg due to copy

---------

Co-authored-by: Comma Device <device@comma.ai>
2024-12-16 16:29:06 -08:00
Harald Schäfer
5018cf75ff North America Model (#34260)
* 0e1c9c12-0472-4a0c-8963-611ad240ec62/400

* rm outputs
2024-12-16 14:17:31 -08:00
Harald Schäfer
f6885dcbec Revert Tinygrad (#34243)
* Revert "dmonitoringmodeld: use cl transform (#34235)"

This reverts commit 684b0b9d4d.

* Revert "load model before calling convert_fp16_to_fp32"

This reverts commit 31606a7d15.

* Revert "bump tinygrad"

This reverts commit 44f58ff758.

* Revert "Tinygrad runner (#34171)"

This reverts commit 7b5a4fbb03.

* Allow init buffer

* typo
2024-12-14 12:15:36 -08:00
ZwX1616
684b0b9d4d dmonitoringmodeld: use cl transform (#34235)
* needs cleanup

* only if tici

* bump tinygrad

* check width

* base modelframe

* .

* need to be args

* more cleanup

* no _frame in base

* tici only

* its DrivingModelFrame

* .6 is fair

---------

Co-authored-by: Comma Device <device@comma.ai>
2024-12-14 00:44:15 -08:00
Harald Schäfer
8743bc4fe2 Null Pointer Model (#34111)
* e8cb7f27-e448-4c15-90c2-ac440cd5a042/400

* 0078ad07-4d46-4086-820f-23d61c90e07f/400

* 4bd74082-70af-47da-8156-e84ebf4d4812/400

* 2a074022-5c2c-4628-97f9-f54849a936a6/400

* 0660aa81-93c5-41b7-9cc2-dc8816a512cd/400

* Clip curvature to reasonable limits

* Better curvature and speed clips

* typo

* typo

* 31aa62c3-b373-4878-8f2e-5107305de187/400

* 384690ca-9b8a-41fe-9bcd-389b20fc6aa4/400

* ref commit

---------

Co-authored-by: Yassine <yassine.y10@gmail.com>
2024-12-13 19:14:21 -08:00
Harald Schäfer
7b5a4fbb03 Tinygrad runner (#34171)
* squash

* bump tg

* bump tg

* debump tinygrad

* bump tinygrad

* bump tg

* Skip init iteration

* fixes

* cleanups

* skip first test sample

* typos

* linter unhappy

* update cpu usage

* OPENCL just zeros for now

* imports

* Try printing

* Runs again, but slower

* unused import

* Allow more buffer with tg and all on gpu

* bump tinygrad

* seems ok

* stricter timings for driving looser for dm

* try llvm

* check nvidia

* More timeout for now

* make test pass

* Revert "try llvm"

This reverts commit ef136e478320101fea262bae3579e558da991902.

* small fixes

* whitespace

* revert test timeout

* No model runners

* Always CPU always fast

* No onnx runtime GPU

* more cores

* cleanup

* Is this faster

* Is this faster

* at least runs

* FP32 is faster than 16

* fix deps

* whitespace

* comment

---------

Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
2024-12-11 23:15:20 -08:00
Adeeb Shihadeh
d9d57e5d6f Revert ISP image processing + tinygrad bump (#34020)
* Revert "Replace ThneedModel with TinygradModel (#33532)"

This reverts commit da952e9b64.

* Revert "camerad: move E + D cams image pipelines to the IFE (#33959)"

This reverts commit f2a1cce42b.
2024-11-13 19:27:11 -08:00
Mitchell Goff
da952e9b64 Replace ThneedModel with TinygradModel (#33532)
* squash

* bump tg

* bump tg

* debump tinygrad

* bump tinygrad

* bump tg

* Skip init iteration

* fixes

* cleanups

* skip first test sample

* typos

* linter unhappy

* update cpu usage

* OPENCL just zeros for now

* imports

* Try printing

* Runs again, but slower

* unused import

* Allow more buffer with tg and all on gpu

* bump tinygrad

---------

Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
Co-authored-by: Bruce Wayne <harald.the.engineer@gmail.com>
2024-11-11 13:32:21 -08:00
Shane Smiskol
7556233cca Reapply "move car.capnp to opendbc (#33722)" (#33728)
* Reapply "move car.capnp to opendbc" (#33725)

This reverts commit 9d52a5b485.

* why can't i repro?!

* Revert "why can't i repro?!"

This reverts commit 0435d218f790faf7b7aaed27d05ab9ee67b087e6.

* does this cause card to try and read it?

* better place

* wtf

* Reapply "why can't i repro?!"

This reverts commit d24fd5a0abf454f47d5591e3b39039fdc4d0251c.

* also here
2024-10-03 22:47:03 -07:00
Shane Smiskol
9d52a5b485 Revert "move car.capnp to opendbc" (#33725)
Revert "move car.capnp to opendbc (#33722)"

This reverts commit 8149f7cb11.
2024-10-03 16:57:45 -07:00
Shane Smiskol
8149f7cb11 move car.capnp to opendbc (#33722)
* move car.capnp to opendbc

* bump

* do card

* fix some more tests

* rm helpers

* format fp

* more

* whoops

* bump

* bump to master
2024-10-03 16:40:24 -07:00
YassineYousfi
48bd5b9f6b modeld: clip vego (#33699) 2024-10-01 21:33:46 -07:00
Harald Schäfer
e442425c9d MLSIM trained model V0 (#33676)
* squash f7343ba0-3f7d-4776-9819-2219b8d2d667/270

* fix broken ref

* Model replay ref commit
2024-10-01 02:58:20 -07:00
Bruce Wayne
d4e257213d Modeld: cannot run without wide img 2024-09-30 00:05:28 -07:00
Shane Smiskol
4f019b5f60 move selfdrive/car to opendbc (#32630)
* move most of /car

* move some car tests

move some car tests

* fix selfdrive/car/tests

* fix selfdrive/controls tests

* fix the rest of the selfdrive tests

* bump opendbc

* fix all tests

* few more non-test references

* remove opcar and move docs to car

fix these debugging scripts

fix docs

* bump opendbc and panda

forgot panda
old-commit-hash: e735a7f379
2024-08-17 00:54:51 -07:00
Shane Smiskol
95224db413 selfdrive/car: ban cereal and capnp (#33208)
* ban cereal and msgq

* common too

* do toyota/values.py

* do all fingerprints

* example without builder

* this still works, but no type checking anymore

* stash

* wtf, how does this work

* okay actually not bad

* safe

* epic!

* stash data_structures.py

* some clean up

* hell yeah

* clean up old file

* add to delete

* delete

This reverts commit 90239b7797ace31ee647a2fdbd67e0c3faa98dcf.

* switch more CarParams stuff over

remove unused

* fix car tests by removing cereal! mypy forgets about dataclass if we wrap it :(

* fix this too

* fix this too

* remove more cereal and add some good hyundai tests

* bunch more typing

* override default with 20hz radar

* temp capnp converter helper

* more lateralTuning

* small union replicator is better than what i was trying, and fixes mypy dynamic typing issues

* can keep all this the same now!

* type ret: CarParams, add more missing structs, revert lateralTuning changes (smaller diff!)

* revert more

* get first enum automatically, but ofc mypy doesn't pick up the new metaclass so can't use :(

would have been `CarParams.NetworkLocation()`

* Revert "get first enum automatically, but ofc mypy doesn't pick up the new metaclass so can't use :("

This reverts commit bb28b228becba932052d2fc5a4389784027435b1.

* remove cereal from car_helpers (TODO: caching)

* remove a bunch of temp lines

* use dataclass_transform!

* remove some car.CarParams from the interfaces

* remove rest of car.CarParams from the interfaces

* same which() API

* sort

* from cereal/cache from fingerprinting!

* more typing

* dataclass to capnp helper for CarParams, cached it since it's kinda slow

* (partial) fix process replay fingerprintig for new API

* latcontrollers take capnp

* forgot this

* fix test_models

* fix unit tests

* not here

* VehicleModel and controller still takes capnp CP since they get it from Params()

* fix modeld test

* more fix

* need to namespace to structs, since CarState is both class and struct

* this was never in the base class?!

* clean that up again

* fix import error

fix import error

* cmts and more structs

* remove some more cereal from toyota + convert CarState to capnp

* bruh this was wrong

* replace more cereal

* EventName is one of the last things...

* replace a bunch more cereal.car

* missing imports

* more

* can fix this typing now

* proper toyota+others CS typing!

* mypy can detect return type of CS.update() now

* fix redeclaration of cruise_buttons type

* mypy is only complaining about events now

* temp fix

* add carControl struct

* replace CarControl

i hope there's no circular imports in hyundai's CC

* fine now

* lol this was wrong too

* fix crash

* include my failed attempts at recursively converting to dataclass (doesn't implicitly convert types/recursively :( )

but attrs does, maybe will switch in the future

* clean up

* try out attr.s for its converter (doesn't work recursively yet, but interesting!)

* Revert "try out attr.s for its converter (doesn't work recursively yet, but interesting!)"

This reverts commit ff2434f7bbd45a4d4bfb21f7d6712d1f1c3bcde9.

* test processes doesn't fail anymore (on toyota)!

* fix honda crash

* stash

* Revert "stash"

This reverts commit c1762af4e776790e4ad1322ad4ce0610157346e0.

* remove a bunch more cereal!

* LET'S GOOO

* fix these tests

* and these

* and that

* stash, something is wrong with hyundai enable

* Revert "stash, something is wrong with hyundai enable"

This reverts commit 39cf327def258e2959fe23cd7a550a858f6d8f03.

* forgot these

* remove cereal from fw_versions

* Revert "remove cereal from fw_versions"

This reverts commit 232b37cd409b55d04b1afc90d4a80c49e710eb56.

* remove rest of the cereal exceptions!

* fix that

* add typing to radard since I didn't realize RI.update() switched from cereal to structs

* and here too!

* add TODO for slots

* needed CS to be capnp, fix comparisons, and type hint car_specific so it's easier to catch type issues (capnp isn't detected by mypy :( )

* remove the struct converter

* save ~4-5% CPU at 100hz, we don't modify after so no need to deepcopy

btw pickle.loads(pickle.dumps()) is faster by ~1% CPU

* deepcopy -> copy: we can technically make a reference, but copy is almost free and less error-prone

saves ~1% CPU

* add non-copying asdict function

* should save ~3% CPU (still 4% above baseline)

* fix that, no dict support

* ~27% decrease in time for 20k iterations on 3X (3.37857 -> 2.4821s)

* give a better name

* fix

* dont support none, capitalize

* sheesh, this called type() on every field

* remove CS.events, clean up

* bump card %

* this was a bug on master!

* add a which enum

* default to pid

* revert

* update refs

* not needed, but consistent

* just Ecu

* don't need to do this in this pr

* clean up

* no cast

* consistent typing

* rm

* fix

* can do this if we're desperate for the last few %

* Revert "can do this if we're desperate for the last few %"

This reverts commit 18e11ac7883a0a56583750b1cc5a2b13011e7299.

* type this

* don't need to convert carControl

* i guess don't support set either

* fix CP type hint

* simplify that
old-commit-hash: 6a15c42143
2024-08-16 15:13:00 -07:00
Adeeb Shihadeh
fc469e5f80 Revert "Remove setproctitle (#32716)" (#33195)
* Revert "Remove `setproctitle` (#32716)"

This reverts commit 9020d1931f4cd72f8162e4402278bdf76e3746f2.

* uv lock

* old name
old-commit-hash: 29b58d4f2f
2024-08-05 16:42:22 -07:00
Dean Lee
45c3992e43 CarParams: set no traversal limit (#32929)
* Increase traversal_limit_in_words to Fix Traversal Limit Errors

* just use log_from_bytes

* come on pycharm

* no free lunch

---------

Co-authored-by: Shane Smiskol <shane@smiskol.com>
old-commit-hash: 12d729a0a0
2024-07-08 13:50:31 -07:00
Kacper Rączy
102188c245 drivingModelData: new model packet for qlogs (#32821)
* Mini model v2 prototype

* Apply suggestions

* Add meta

* Support for new packet in model

* Support in process replay

* Meta

* Add it to services

* Update model replay

* Add drivingModelData in model_replay

* Update ref

* MAX_FRAMES*3

* No modelv2, but drivingModelData at 2hz

* Polypath

* Dont use keyword

* Update model ref commit

* xyz coeff

* Fix field name

* Update ref commit

* Min qlog size 0.6

---------

Co-authored-by: Comma Device <device@comma.ai>
old-commit-hash: 2059f986df
2024-06-28 19:27:58 -07:00
Adeeb Shihadeh
9cfdd6200d Remove setproctitle (#32716)
* Custom setproctitle (#32667)

* add custom setproctitle

* add test

* Update poetry.lock

* fix lint

* support only Linux

* test only Linux

* final lint

* Update test_setproctitle.py

* Update setproctitle.py

* convert to threadnames

* delete proctitles

* Check str len and use PR_GET_NAME

* fix poetry.lock

* lint fix

* Update common/threadname.py

---------

Co-authored-by: reddyn12 <nikhilr.ssm@gmail.com>
Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>

* revert that for now

* use last 15

* fix

* use name

* update those

* and modeld

* rm

---------

Co-authored-by: schlimeszn <138847413+schlimeszn@users.noreply.github.com>
Co-authored-by: reddyn12 <nikhilr.ssm@gmail.com>
Co-authored-by: Comma Device <device@comma.ai>
old-commit-hash: 83ac80c6c8
2024-06-11 21:57:32 -07:00
Adeeb Shihadeh
e920e57530 Revert "Custom setproctitle (#32667)"
This reverts commit 1f45e163559baa2a6127d5a5e8deb7067e32f813.

old-commit-hash: 5b51f03967
2024-06-10 19:55:41 -07:00
schlimeszn
afea57ae2a Custom setproctitle (#32667)
* add custom setproctitle

* add test

* Update poetry.lock

* fix lint

* support only Linux

* test only Linux

* final lint

* Update test_setproctitle.py

* Update setproctitle.py

* convert to threadnames

* delete proctitles

* Check str len and use PR_GET_NAME

* fix poetry.lock

* lint fix

* Update common/threadname.py

---------

Co-authored-by: reddyn12 <nikhilr.ssm@gmail.com>
Co-authored-by: Adeeb Shihadeh <adeebshihadeh@gmail.com>
old-commit-hash: 3365ed5eff
2024-06-10 18:59:01 -07:00