Merge branch 'master-new' into mads-new

.github/PULL_REQUEST_TEMPLATE/car_port.md

@@ -8,7 +8,7 @@ assignees: ''
 
 **Checklist**
 
-- [ ] added entry to CAR in selfdrive/car/*/values.py and ran `selfdrive/opcar/docs.py` to generate new docs
+- [ ] added entry to CAR in selfdrive/car/*/values.py and ran `selfdrive/car/docs.py` to generate new docs
 - [ ] test route added to [routes.py](https://github.com/commaai/openpilot/blob/master/selfdrive/car/tests/routes.py)
 - [ ] route with openpilot:
 - [ ] route with stock system:

.github/pull_request_template.md

@@ -44,7 +44,7 @@ Explain how you tested this bug fix.
 
 **Checklist**
 
-- [ ] added entry to CAR in selfdrive/car/*/values.py and ran `selfdrive/opcar/docs.py` to generate new docs
+- [ ] added entry to CAR in selfdrive/car/*/values.py and ran `selfdrive/car/docs.py` to generate new docs
 - [ ] test route added to [routes.py](https://github.com/commaai/openpilot/blob/master/selfdrive/car/tests/routes.py)
 - [ ] route with openpilot:
 - [ ] route with stock system:

.github/workflows/ci_weekly_run.yaml

@@ -15,7 +15,3 @@ jobs:
     uses: sunnypilot/sunnypilot/.github/workflows/selfdrive_tests.yaml@master
     with:
       run_number: ${{ inputs.run_number }}
-  tools_tests:
-    uses: sunnypilot/sunnypilot/.github/workflows/tools_tests.yaml@master
-    with:
-      run_number: ${{ inputs.run_number }}

.github/workflows/docs.yaml

@@ -18,7 +18,7 @@ concurrency:
 jobs:
   docs:
     name: build docs
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-24.04
     steps:
     - uses: commaai/timeout@v1
 

.github/workflows/selfdrive_tests.yaml

@@ -239,7 +239,7 @@ jobs:
       uses: actions/cache@v4
       with:
         path: .ci_cache/comma_download_cache
-        key: car_models-${{ hashFiles('selfdrive/car/tests/test_models.py', 'selfdrive/car/tests/routes.py') }}-${{ matrix.job }}
+        key: car_models-${{ hashFiles('selfdrive/car/tests/test_models.py', 'opendbc/car/tests/routes.py') }}-${{ matrix.job }}
     - name: Build openpilot
       run: ${{ env.RUN }} "scons -j$(nproc)"
     - name: Test car models
@@ -317,6 +317,7 @@ jobs:
     runs-on:
       - ${{ ((github.repository == 'commaai/openpilot') && ((github.event_name != 'pull_request') || (github.event.pull_request.head.repo.full_name == 'commaai/openpilot'))) && 'namespace-profile-amd64-8x16' || 'ubuntu-24.04' }}
       - ${{ ((github.repository == 'commaai/openpilot') && ((github.event_name != 'pull_request') || (github.event.pull_request.head.repo.full_name == 'commaai/openpilot'))) && 'namespace-experiments:docker.builds.local-cache=separate' || 'ubuntu-24.04' }}
+    if: (github.repository == 'commaai/openpilot') && ((github.event_name != 'pull_request') || (github.event.pull_request.head.repo.full_name == 'commaai/openpilot'))
     steps:
     - uses: actions/checkout@v4
       with:

Dockerfile.openpilot

@@ -1,9 +1,9 @@
 FROM ghcr.io/commaai/openpilot-base:latest
 
-ENV PYTHONUNBUFFERED 1
+ENV PYTHONUNBUFFERED=1
 
-ENV OPENPILOT_PATH /home/batman/openpilot
-ENV PYTHONPATH ${OPENPILOT_PATH}:${PYTHONPATH}
+ENV OPENPILOT_PATH=/home/batman/openpilot
+ENV PYTHONPATH=${OPENPILOT_PATH}:${PYTHONPATH}
 
 RUN mkdir -p ${OPENPILOT_PATH}
 WORKDIR ${OPENPILOT_PATH}

Dockerfile.openpilot_base

@@ -1,6 +1,6 @@
 FROM ubuntu:24.04
 
-ENV PYTHONUNBUFFERED 1
+ENV PYTHONUNBUFFERED=1
 
 ENV DEBIAN_FRONTEND=noninteractive
 RUN apt-get update && \
@@ -8,9 +8,9 @@ RUN apt-get update && \
     rm -rf /var/lib/apt/lists/*
 
 RUN sed -i -e 's/# en_US.UTF-8 UTF-8/en_US.UTF-8 UTF-8/' /etc/locale.gen && locale-gen
-ENV LANG en_US.UTF-8
-ENV LANGUAGE en_US:en
-ENV LC_ALL en_US.UTF-8
+ENV LANG=en_US.UTF-8
+ENV LANGUAGE=en_US:en
+ENV LC_ALL=en_US.UTF-8
 
 COPY tools/install_ubuntu_dependencies.sh /tmp/tools/
 RUN /tmp/tools/install_ubuntu_dependencies.sh && \
@@ -55,9 +55,9 @@ RUN mkdir -p /tmp/opencl-driver-intel && \
     cd / && \
     rm -rf /tmp/opencl-driver-intel
 
-ENV NVIDIA_VISIBLE_DEVICES all
-ENV NVIDIA_DRIVER_CAPABILITIES graphics,utility,compute
-ENV QTWEBENGINE_DISABLE_SANDBOX 1
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV NVIDIA_DRIVER_CAPABILITIES=graphics,utility,compute
+ENV QTWEBENGINE_DISABLE_SANDBOX=1
 
 RUN dbus-uuidgen > /etc/machine-id
 

Jenkinsfile

@@ -79,6 +79,10 @@ def deviceStage(String stageName, String deviceType, List extra_env, def steps)
         return
     }
 
+    if (isReplay()) {
+      error("REPLAYING TESTS IS NOT ALLOWED. FIX THEM INSTEAD.")
+    }
+
     def extra = extra_env.collect { "export ${it}" }.join('\n');
     def branch = env.BRANCH_NAME ?: 'master';
     def gitDiff = sh returnStdout: true, script: 'curl -s -H "Authorization: Bearer ${GITHUB_COMMENTS_TOKEN}" https://api.github.com/repos/commaai/openpilot/compare/master...${GIT_BRANCH} | jq .files[].filename || echo "/"', label: 'Getting changes'
@@ -123,6 +127,11 @@ def hasPathChanged(String gitDiff, List<String> paths) {
   return false
 }
 
+def isReplay() {
+  def replayClass = "org.jenkinsci.plugins.workflow.cps.replay.ReplayCause"
+  return currentBuild.rawBuild.getCauses().any{ cause -> cause.toString().contains(replayClass) }
+}
+
 def setupCredentials() {
   withCredentials([
     string(credentialsId: 'azure_token', variable: 'AZURE_TOKEN'),

README.md

@@ -38,7 +38,8 @@ Quick start: `bash <(curl -fsSL openpilot.comma.ai)`
   </tr>
 </table>
 
-To start using openpilot in a car
+
+Using openpilot in a car
 ------
 
 To use openpilot in a car, you need four things:
@@ -49,6 +50,14 @@ To use openpilot in a car, you need four things:
 
 We have detailed instructions for [how to install the harness and device in a car](https://comma.ai/setup). Note that it's possible to run openpilot on [other hardware](https://blog.comma.ai/self-driving-car-for-free/), although it's not plug-and-play.
 
+### Branches
+| branch           | URL                                    | description                                                                         |
+|------------------|----------------------------------------|-------------------------------------------------------------------------------------|
+| `release3`         | openilot.comma.ai                      | This is openpilot's release branch.                                                 |
+| `release3-staging` | openpilot-test.comma.ai                | This is the staging branch for releases. Use it to get new releases slightly early. |
+| `nightly`          | openpilot-nightly.comma.ai             | This is the bleeding edge development branch. Do not expect this to be stable.      |
+| `nightly-dev`      | installer.comma.ai/commaai/nightly-dev | Same as nightly, but includes experimental development features for some cars.      |
+
 To start developing openpilot
 ------
 

cereal/log.capnp

@@ -2516,6 +2516,14 @@ struct Microphone {
   filteredSoundPressureWeightedDb @2 :Float32;
 }
 
+struct Touch {
+  sec @0 :Int64;
+  usec @1 :Int64;
+  type @2 :UInt8;
+  code @3 :Int32;
+  value @4 :Int32;
+}
+
 struct Event {
   logMonoTime @0 :UInt64;  # nanoseconds
   valid @67 :Bool = true;
@@ -2596,6 +2604,9 @@ struct Event {
     logMessage @18 :Text;
     errorLogMessage @85 :Text;
 
+    # touch frame
+    touch @135 :List(Touch);
+
     # navigation
     navInstruction @82 :NavInstruction;
     navRoute @83 :NavRoute;

cereal/services.py

@@ -22,6 +22,7 @@ _services: dict[str, tuple] = {
   "temperatureSensor2": (True, 2., 200),
   "gpsNMEA": (True, 9.),
   "deviceState": (True, 2., 1),
+  "touch": (True, 20., 1),
   "can": (True, 100., 2053),  # decimation gives ~3 msgs in a full segment
   "controlsState": (True, 100., 10),
   "selfdriveState": (True, 100., 10),

common/transformations/model.py

@@ -1,7 +1,7 @@
 import numpy as np
 
 from openpilot.common.transformations.orientation import rot_from_euler
-from openpilot.common.transformations.camera import get_view_frame_from_calib_frame, view_frame_from_device_frame
+from openpilot.common.transformations.camera import get_view_frame_from_calib_frame, view_frame_from_device_frame, _ar_ox_fisheye
 
 # segnet
 SEGNET_SIZE = (512, 384)
@@ -39,6 +39,13 @@ sbigmodel_intrinsics = np.array([
   [0.0,  sbigmodel_fl,      0.5 * (256 + MEDMODEL_CY)],
   [0.0,  0.0,                                     1.0]])
 
+DM_INPUT_SIZE = (1440, 960)
+dmonitoringmodel_fl = _ar_ox_fisheye.focal_length
+dmonitoringmodel_intrinsics = np.array([
+  [dmonitoringmodel_fl,  0.0, DM_INPUT_SIZE[0]/2],
+  [0.0, dmonitoringmodel_fl, DM_INPUT_SIZE[1]/2 - (_ar_ox_fisheye.height - DM_INPUT_SIZE[1])/2],
+  [0.0,  0.0, 1.0]])
+
 bigmodel_frame_from_calib_frame = np.dot(bigmodel_intrinsics,
   get_view_frame_from_calib_frame(0, 0, 0, 0))
 

docs/CARS.md

@@ -103,7 +103,7 @@ A supported vehicle is one that just works when you install a comma device. All
 |Hyundai|Ioniq Plug-in Hybrid 2020-22|All|openpilot available[<sup>1</sup>](#footnotes)|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai H connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Ioniq Plug-in Hybrid 2020-22">Buy Here</a></sub></details>||
 |Hyundai|Kona 2020|Smart Cruise Control (SCC)|openpilot available[<sup>1</sup>](#footnotes)|6 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-empty.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai B connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Kona 2020">Buy Here</a></sub></details>||
 |Hyundai|Kona Electric 2018-21|Smart Cruise Control (SCC)|openpilot available[<sup>1</sup>](#footnotes)|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai G connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Kona Electric 2018-21">Buy Here</a></sub></details>||
-|Hyundai|Kona Electric 2022-23|Smart Cruise Control (SCC)|Stock|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai O connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Kona Electric 2022-23">Buy Here</a></sub></details>||
+|Hyundai|Kona Electric 2022-23|Smart Cruise Control (SCC)|openpilot available[<sup>1</sup>](#footnotes)|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai O connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Kona Electric 2022-23">Buy Here</a></sub></details>||
 |Hyundai|Kona Electric (with HDA II, Korea only) 2023[<sup>5</sup>](#footnotes)|Smart Cruise Control (SCC)|Stock|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai R connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Kona Electric (with HDA II, Korea only) 2023">Buy Here</a></sub></details>|<a href="https://www.youtube.com/watch?v=U2fOCmcQ8hw" target="_blank"><img height="18px" src="assets/icon-youtube.svg"></img></a>|
 |Hyundai|Kona Hybrid 2020|Smart Cruise Control (SCC)|openpilot available[<sup>1</sup>](#footnotes)|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai I connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Kona Hybrid 2020">Buy Here</a></sub></details>||
 |Hyundai|Palisade 2020-22|All|openpilot available[<sup>1</sup>](#footnotes)|0 mph|0 mph|[![star](assets/icon-star-full.svg)](##)|[![star](assets/icon-star-full.svg)](##)|<details><summary>Parts</summary><sub>- 1 Hyundai H connector<br>- 1 RJ45 cable (7 ft)<br>- 1 comma 3X<br>- 1 comma power v2<br>- 1 harness box<br>- 1 mount<br>- 1 right angle OBD-C cable (1.5 ft)<br><a href="https://comma.ai/shop/comma-3x.html?make=Hyundai&model=Palisade 2020-22">Buy Here</a></sub></details>|<a href="https://youtu.be/TAnDqjF4fDY?t=456" target="_blank"><img height="18px" src="assets/icon-youtube.svg"></img></a>|

docs/css/tooltip.css

@@ -0,0 +1,44 @@
+[data-tooltip] {
+  position: relative;
+  display: inline-block;
+  border-bottom: 1px dotted black;
+}
+
+[data-tooltip] .tooltip-content {
+  width: max-content;
+  max-width: 25em;
+  position: absolute;
+  top: 100%;
+  left: 50%;
+  transform: translateX(-50%);
+  background-color: white;
+  color: #404040;
+  box-shadow: 0 4px 14px 0 rgba(0,0,0,.2), 0 0 0 1px rgba(0,0,0,.05);
+  padding: 10px;
+  font: 14px/1.5 Lato, proxima-nova, Helvetica Neue, Arial, sans-serif;
+  text-decoration: none;
+  opacity: 0;
+  visibility: hidden;
+  transition: opacity 0.1s, visibility 0s;
+  z-index: 1000;
+  pointer-events: none; /* Prevent accidental interaction */
+}
+
+[data-tooltip]:hover .tooltip-content {
+  opacity: 1;
+  visibility: visible;
+  pointer-events: auto; /* Allow interaction when visible */
+}
+
+.tooltip-content .tooltip-glossary-link {
+  display: inline-block;
+  margin-top: 8px;
+  font-size: 12px;
+  color: #007bff;
+  text-decoration: none;
+}
+
+.tooltip-content .tooltip-glossary-link:hover {
+  color: #0056b3;
+  text-decoration: underline;
+}

docs/hooks/glossary.py

@@ -0,0 +1,68 @@
+import re
+import tomllib
+
+def load_glossary(file_path="docs/glossary.toml"):
+  with open(file_path, "rb") as f:
+    glossary_data = tomllib.load(f)
+  return glossary_data.get("glossary", {})
+
+def generate_anchor_id(name):
+  return name.replace(" ", "-").replace("_", "-").lower()
+
+def format_markdown_term(name, definition):
+  anchor_id = generate_anchor_id(name)
+  markdown = f"* [**{name.replace('_', ' ').title()}**](#{anchor_id})"
+  if definition.get("abbreviation"):
+    markdown += f" *({definition['abbreviation']})*"
+  if definition.get("description"):
+    markdown += f": {definition['description']}\n"
+  return markdown
+
+def glossary_markdown(vocabulary):
+  markdown = ""
+  for category, terms in vocabulary.items():
+    markdown += f"## {category.replace('_', ' ').title()}\n\n"
+    for name, definition in terms.items():
+      markdown += format_markdown_term(name, definition)
+  return markdown
+
+def format_tooltip_html(term_key, definition, html):
+  display_term = term_key.replace("_", " ").title()
+  clean_description = re.sub(r"\[(.+)]\(.+\)", r"\1", definition["description"])
+  glossary_link = (
+    f"<a href='/concepts/glossary#{term_key}' class='tooltip-glossary-link' title='View in glossary'>Glossary🔗</a>"
+  )
+  return re.sub(
+    re.escape(display_term),
+    lambda
+    match: f"<span data-tooltip>{match.group(0)}<span class='tooltip-content'>{clean_description} {glossary_link}</span></span>",
+    html,
+    flags=re.IGNORECASE,
+  )
+
+def apply_tooltip(_term_key, _definition, pattern, html):
+  return re.sub(
+    pattern,
+    lambda match: format_tooltip_html(_term_key, _definition, match.group(0)),
+    html,
+    flags=re.IGNORECASE,
+  )
+
+def tooltip_html(vocabulary, html):
+  for _category, terms in vocabulary.items():
+    for term_key, definition in terms.items():
+      if definition.get("description"):
+        pattern = rf"(?<!\w){re.escape(term_key.replace('_', ' ').title())}(?![^<]*<\/a>)(?!\([^)]*\))"
+        html = apply_tooltip(term_key, definition, pattern, html)
+  return html
+
+# Page Hooks
+def on_page_markdown(markdown, **kwargs):
+  glossary = load_glossary()
+  return markdown.replace("{{GLOSSARY_DEFINITIONS}}", glossary_markdown(glossary))
+
+def on_page_content(html, **kwargs):
+  if kwargs.get("page").title == "Glossary":
+    return html
+  glossary = load_glossary()
+  return tooltip_html(glossary, html)

launch_env.sh

@@ -7,7 +7,7 @@ export OPENBLAS_NUM_THREADS=1
 export VECLIB_MAXIMUM_THREADS=1
 
 if [ -z "$AGNOS_VERSION" ]; then
-  export AGNOS_VERSION="11.3"
+  export AGNOS_VERSION="11.4"
 fi
 
 export STAGING_ROOT="/data/safe_staging"

mkdocs.yml

@@ -8,6 +8,10 @@ strict: true
 docs_dir: docs
 site_dir: docs_site/
 
+hooks:
+  - docs/hooks/glossary.py
+extra_css:
+  - css/tooltip.css
 theme:
   name: readthedocs
   navigation_depth: 3

selfdrive/car/car_specific.py

@@ -148,7 +148,8 @@ class CarSpecificEvents:
       # To avoid re-engaging when openpilot cancels, check user engagement intention via buttons
       # Main button also can trigger an engagement on these cars
       self.cruise_buttons.append(any(ev.type in HYUNDAI_ENABLE_BUTTONS for ev in CS.buttonEvents))
-      events = self.create_common_events(CS, CS_prev, pcm_enable=self.CP.pcmCruise, allow_enable=any(self.cruise_buttons))
+      events = self.create_common_events(CS, CS_prev, extra_gears=(GearShifter.sport, GearShifter.manumatic),
+                                         pcm_enable=self.CP.pcmCruise, allow_enable=any(self.cruise_buttons))
 
       # low speed steer alert hysteresis logic (only for cars with steer cut off above 10 m/s)
       if CS.vEgo < (self.CP.minSteerSpeed + 2.) and self.CP.minSteerSpeed > 10.:

selfdrive/car/tests/test_docs.py

@@ -17,7 +17,7 @@ class TestCarDocs:
     with open(CARS_MD_OUT) as f:
       current_cars_md = f.read()
 
-    assert generated_cars_md == current_cars_md, "Run selfdrive/opcar/docs.py to update the compatibility documentation"
+    assert generated_cars_md == current_cars_md, "Run selfdrive/car/docs.py to update the compatibility documentation"
 
   def test_docs_diff(self):
     dump_path = os.path.join(BASEDIR, "selfdrive", "car", "tests", "cars_dump")

selfdrive/controls/lib/drive_helpers.py

@@ -5,12 +5,15 @@ from openpilot.common.realtime import DT_CTRL
 MIN_SPEED = 1.0
 CONTROL_N = 17
 CAR_ROTATION_RADIUS = 0.0
+# This is a turn radius smaller than most cars can achieve
+MAX_CURVATURE = 0.2
 
 # EU guidelines
 MAX_LATERAL_JERK = 5.0
 MAX_VEL_ERR = 5.0
 
 def clip_curvature(v_ego, prev_curvature, new_curvature):
+  new_curvature = clip(new_curvature, -MAX_CURVATURE, MAX_CURVATURE)
   v_ego = max(MIN_SPEED, v_ego)
   max_curvature_rate = MAX_LATERAL_JERK / (v_ego**2) # inexact calculation, check https://github.com/commaai/openpilot/pull/24755
   safe_desired_curvature = clip(new_curvature,

selfdrive/debug/touch_replay.py

@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+import argparse
+
+import numpy as np
+import matplotlib.pyplot as plt
+
+from openpilot.tools.lib.logreader import LogReader
+
+if __name__ == '__main__':
+  parser = argparse.ArgumentParser()
+  parser.add_argument('--width', default=2160, type=int)
+  parser.add_argument('--height', default=1080, type=int)
+  parser.add_argument('--route', default='rlog', type=str)
+  args = parser.parse_args()
+
+  w = args.width
+  h = args.height
+  route = args.route
+
+  fingers = [[-1, -1]] * 5
+  touch_points = []
+  current_slot = 0
+
+  lr = list(LogReader(route))
+  for msg in lr:
+    if msg.which() == 'touch':
+      for event in msg.touch:
+        if event.type == 3 and event.code == 47:
+          current_slot = event.value
+        elif event.type == 3 and event.code == 57 and event.value == -1:
+          fingers[current_slot] = [-1, -1]
+        elif event.type == 3 and event.code == 53:
+          fingers[current_slot][1] = h - (h - event.value)
+          if fingers[current_slot][0] != -1:
+            touch_points.append(fingers[current_slot].copy())
+        elif event.type == 3 and event.code == 54:
+          fingers[current_slot][0] = w - event.value
+          if fingers[current_slot][1] != -1:
+            touch_points.append(fingers[current_slot].copy())
+
+  unique_points, counts = np.unique(touch_points, axis=0, return_counts=True)
+
+  plt.figure(figsize=(10, 3))
+  plt.scatter(unique_points[:, 0], unique_points[:, 1], c=counts, s=counts * 20, edgecolors='red')
+  plt.colorbar()
+  plt.title(f'Touches for {route}')
+  plt.grid(True)
+  plt.show()

selfdrive/modeld/dmonitoringmodeld.py

@@ -13,13 +13,13 @@ from cereal.messaging import PubMaster, SubMaster
 from msgq.visionipc import VisionIpcClient, VisionStreamType, VisionBuf
 from openpilot.common.swaglog import cloudlog
 from openpilot.common.realtime import set_realtime_priority
+from openpilot.common.transformations.model import dmonitoringmodel_intrinsics
+from openpilot.common.transformations.camera import _ar_ox_fisheye, _os_fisheye
+from openpilot.selfdrive.modeld.models.commonmodel_pyx import CLContext, MonitoringModelFrame
 from openpilot.selfdrive.modeld.runners import ModelRunner, Runtime
-from openpilot.selfdrive.modeld.models.commonmodel_pyx import CLContext
 from openpilot.selfdrive.modeld.parse_model_outputs import sigmoid
 
 CALIB_LEN = 3
-MODEL_WIDTH = 1440
-MODEL_HEIGHT = 960
 FEATURE_LEN = 512
 OUTPUT_SIZE = 84 + FEATURE_LEN
 
@@ -62,25 +62,21 @@ class ModelState:
 
   def __init__(self, cl_ctx):
     assert ctypes.sizeof(DMonitoringModelResult) == OUTPUT_SIZE * ctypes.sizeof(ctypes.c_float)
+
+    self.frame = MonitoringModelFrame(cl_ctx)
     self.output = np.zeros(OUTPUT_SIZE, dtype=np.float32)
     self.inputs = {
-      'input_img': np.zeros(MODEL_HEIGHT * MODEL_WIDTH, dtype=np.uint8),
       'calib': np.zeros(CALIB_LEN, dtype=np.float32)}
 
     self.model = ModelRunner(MODEL_PATHS, self.output, Runtime.GPU, False, cl_ctx)
     self.model.addInput("input_img", None)
     self.model.addInput("calib", self.inputs['calib'])
 
-  def run(self, buf:VisionBuf, calib:np.ndarray) -> tuple[np.ndarray, float]:
+  def run(self, buf:VisionBuf, calib:np.ndarray, transform:np.ndarray) -> tuple[np.ndarray, float]:
     self.inputs['calib'][:] = calib
 
-    v_offset = buf.height - MODEL_HEIGHT
-    h_offset = (buf.width - MODEL_WIDTH) // 2
-    buf_data = buf.data.reshape(-1, buf.stride)
-    input_data = self.inputs['input_img'].reshape(MODEL_HEIGHT, MODEL_WIDTH)
-    input_data[:] = buf_data[v_offset:v_offset+MODEL_HEIGHT, h_offset:h_offset+MODEL_WIDTH]
+    self.model.setInputBuffer("input_img", self.frame.prepare(buf, transform.flatten(), None).view(np.float32))
 
-    self.model.setInputBuffer("input_img", self.inputs['input_img'].view(np.float32))
     t1 = time.perf_counter()
     self.model.execute()
     t2 = time.perf_counter()
@@ -137,18 +133,23 @@ def main():
   pm = PubMaster(["driverStateV2"])
 
   calib = np.zeros(CALIB_LEN, dtype=np.float32)
+  model_transform = None
 
   while True:
     buf = vipc_client.recv()
     if buf is None:
       continue
 
+    if model_transform is None:
+      cam = _os_fisheye if buf.width == _os_fisheye.width else _ar_ox_fisheye
+      model_transform = np.linalg.inv(np.dot(dmonitoringmodel_intrinsics, np.linalg.inv(cam.intrinsics))).astype(np.float32)
+
     sm.update(0)
     if sm.updated["liveCalibration"]:
       calib[:] = np.array(sm["liveCalibration"].rpyCalib)
 
     t1 = time.perf_counter()
-    model_output, gpu_execution_time = model.run(buf, calib)
+    model_output, gpu_execution_time = model.run(buf, calib, model_transform)
     t2 = time.perf_counter()
 
     pm.send("driverStateV2", get_driverstate_packet(model_output, vipc_client.frame_id, vipc_client.timestamp_sof, t2 - t1, gpu_execution_time))

selfdrive/modeld/fill_model_msg.py

@@ -3,11 +3,22 @@ import capnp
 import numpy as np
 from cereal import log
 from openpilot.selfdrive.modeld.constants import ModelConstants, Plan, Meta
+from openpilot.selfdrive.controls.lib.drive_helpers import MIN_SPEED
 
 SEND_RAW_PRED = os.getenv('SEND_RAW_PRED')
 
 ConfidenceClass = log.ModelDataV2.ConfidenceClass
 
+def curv_from_psis(psi_target, psi_rate, vego, delay):
+  vego = np.clip(vego, MIN_SPEED, np.inf)
+  curv_from_psi = psi_target / (vego * delay)  # epsilon to prevent divide-by-zero
+  return 2*curv_from_psi - psi_rate / vego
+
+def get_curvature_from_plan(plan, vego, delay):
+  psi_target = np.interp(delay, ModelConstants.T_IDXS, plan[:, Plan.T_FROM_CURRENT_EULER][:, 2])
+  psi_rate = plan[:, Plan.ORIENTATION_RATE][0, 2]
+  return curv_from_psis(psi_target, psi_rate, vego, delay)
+
 class PublishState:
   def __init__(self):
     self.disengage_buffer = np.zeros(ModelConstants.CONFIDENCE_BUFFER_LEN*ModelConstants.DISENGAGE_WIDTH, dtype=np.float32)
@@ -55,14 +66,17 @@ def fill_lane_line_meta(builder, lane_lines, lane_line_probs):
   builder.rightProb = lane_line_probs[2]
 
 def fill_model_msg(base_msg: capnp._DynamicStructBuilder, extended_msg: capnp._DynamicStructBuilder,
-                   net_output_data: dict[str, np.ndarray], publish_state: PublishState,
-                   vipc_frame_id: int, vipc_frame_id_extra: int, frame_id: int, frame_drop: float,
-                   timestamp_eof: int, model_execution_time: float, valid: bool) -> None:
+                   net_output_data: dict[str, np.ndarray], v_ego: float, delay: float,
+                   publish_state: PublishState, vipc_frame_id: int, vipc_frame_id_extra: int,
+                   frame_id: int, frame_drop: float, timestamp_eof: int, model_execution_time: float,
+                   valid: bool) -> None:
   frame_age = frame_id - vipc_frame_id if frame_id > vipc_frame_id else 0
   frame_drop_perc = frame_drop * 100
   extended_msg.valid = valid
   base_msg.valid = valid
 
+  desired_curv = float(get_curvature_from_plan(net_output_data['plan'][0], v_ego, delay))
+
   driving_model_data = base_msg.drivingModelData
 
   driving_model_data.frameId = vipc_frame_id
@@ -71,7 +85,7 @@ def fill_model_msg(base_msg: capnp._DynamicStructBuilder, extended_msg: capnp._D
   driving_model_data.modelExecutionTime = model_execution_time
 
   action = driving_model_data.action
-  action.desiredCurvature = float(net_output_data['desired_curvature'][0,0])
+  action.desiredCurvature = desired_curv
 
   modelV2 = extended_msg.modelV2
   modelV2.frameId = vipc_frame_id
@@ -106,7 +120,7 @@ def fill_model_msg(base_msg: capnp._DynamicStructBuilder, extended_msg: capnp._D
 
   # lateral planning
   action = modelV2.action
-  action.desiredCurvature = float(net_output_data['desired_curvature'][0,0])
+  action.desiredCurvature = desired_curv
 
   # times at X_IDXS according to model plan
   PLAN_T_IDXS = [np.nan] * ModelConstants.IDX_N

selfdrive/modeld/modeld.py

@@ -22,7 +22,7 @@ from openpilot.selfdrive.modeld.runners import ModelRunner, Runtime
 from openpilot.selfdrive.modeld.parse_model_outputs import Parser
 from openpilot.selfdrive.modeld.fill_model_msg import fill_model_msg, fill_pose_msg, PublishState
 from openpilot.selfdrive.modeld.constants import ModelConstants
-from openpilot.selfdrive.modeld.models.commonmodel_pyx import ModelFrame, CLContext
+from openpilot.selfdrive.modeld.models.commonmodel_pyx import DrivingModelFrame, CLContext
 
 PROCESS_NAME = "selfdrive.modeld.modeld"
 SEND_RAW_PRED = os.getenv('SEND_RAW_PRED')
@@ -44,27 +44,24 @@ class FrameMeta:
       self.frame_id, self.timestamp_sof, self.timestamp_eof = vipc.frame_id, vipc.timestamp_sof, vipc.timestamp_eof
 
 class ModelState:
-  frame: ModelFrame
-  wide_frame: ModelFrame
+  frame: DrivingModelFrame
+  wide_frame: DrivingModelFrame
   inputs: dict[str, np.ndarray]
   output: np.ndarray
   prev_desire: np.ndarray  # for tracking the rising edge of the pulse
   model: ModelRunner
 
   def __init__(self, context: CLContext):
-    self.frame = ModelFrame(context)
-    self.wide_frame = ModelFrame(context)
+    self.frame = DrivingModelFrame(context)
+    self.wide_frame = DrivingModelFrame(context)
     self.prev_desire = np.zeros(ModelConstants.DESIRE_LEN, dtype=np.float32)
     self.full_features_20Hz = np.zeros((ModelConstants.FULL_HISTORY_BUFFER_LEN, ModelConstants.FEATURE_LEN), dtype=np.float32)
     self.desire_20Hz =  np.zeros((ModelConstants.FULL_HISTORY_BUFFER_LEN + 1, ModelConstants.DESIRE_LEN), dtype=np.float32)
-    self.prev_desired_curv_20hz = np.zeros((ModelConstants.FULL_HISTORY_BUFFER_LEN + 1, ModelConstants.PREV_DESIRED_CURV_LEN), dtype=np.float32)
 
     # img buffers are managed in openCL transform code
     self.inputs = {
       'desire': np.zeros(ModelConstants.DESIRE_LEN * (ModelConstants.HISTORY_BUFFER_LEN+1), dtype=np.float32),
       'traffic_convention': np.zeros(ModelConstants.TRAFFIC_CONVENTION_LEN, dtype=np.float32),
-      'lateral_control_params': np.zeros(ModelConstants.LATERAL_CONTROL_PARAMS_LEN, dtype=np.float32),
-      'prev_desired_curv': np.zeros(ModelConstants.PREV_DESIRED_CURV_LEN * (ModelConstants.HISTORY_BUFFER_LEN+1), dtype=np.float32),
       'features_buffer': np.zeros(ModelConstants.HISTORY_BUFFER_LEN * ModelConstants.FEATURE_LEN, dtype=np.float32),
     }
 
@@ -100,7 +97,6 @@ class ModelState:
     self.inputs['desire'][:] = self.desire_20Hz.reshape((25,4,-1)).max(axis=1).flatten()
 
     self.inputs['traffic_convention'][:] = inputs['traffic_convention']
-    self.inputs['lateral_control_params'][:] = inputs['lateral_control_params']
 
     self.model.setInputBuffer("input_imgs", self.frame.prepare(buf, transform.flatten(), self.model.getCLBuffer("input_imgs")))
     self.model.setInputBuffer("big_input_imgs", self.wide_frame.prepare(wbuf, transform_wide.flatten(), self.model.getCLBuffer("big_input_imgs")))
@@ -114,13 +110,8 @@ class ModelState:
     self.full_features_20Hz[:-1] = self.full_features_20Hz[1:]
     self.full_features_20Hz[-1] = outputs['hidden_state'][0, :]
 
-    self.prev_desired_curv_20hz[:-1] = self.prev_desired_curv_20hz[1:]
-    self.prev_desired_curv_20hz[-1] = outputs['desired_curvature'][0, :]
-
     idxs = np.arange(-4,-100,-4)[::-1]
     self.inputs['features_buffer'][:] = self.full_features_20Hz[idxs].flatten()
-    # TODO model only uses last value now, once that changes we need to input strided action history buffer
-    self.inputs['prev_desired_curv'][-ModelConstants.PREV_DESIRED_CURV_LEN:] = 0. * self.prev_desired_curv_20hz[-4, :]
     return outputs
 
 
@@ -231,7 +222,6 @@ def main(demo=False):
     is_rhd = sm["driverMonitoringState"].isRHD
     frame_id = sm["roadCameraState"].frameId
     v_ego = max(sm["carState"].vEgo, 0.)
-    lateral_control_params = np.array([v_ego, steer_delay], dtype=np.float32)
     if sm.updated["liveCalibration"] and sm.seen['roadCameraState'] and sm.seen['deviceState']:
       device_from_calib_euler = np.array(sm["liveCalibration"].rpyCalib, dtype=np.float32)
       dc = DEVICE_CAMERAS[(str(sm['deviceState'].deviceType), str(sm['roadCameraState'].sensor))]
@@ -262,7 +252,6 @@ def main(demo=False):
     inputs:dict[str, np.ndarray] = {
       'desire': vec_desire,
       'traffic_convention': traffic_convention,
-      'lateral_control_params': lateral_control_params,
       }
 
     mt1 = time.perf_counter()
@@ -274,7 +263,8 @@ def main(demo=False):
       modelv2_send = messaging.new_message('modelV2')
       drivingdata_send = messaging.new_message('drivingModelData')
       posenet_send = messaging.new_message('cameraOdometry')
-      fill_model_msg(drivingdata_send, modelv2_send, model_output, publish_state, meta_main.frame_id, meta_extra.frame_id, frame_id,
+      fill_model_msg(drivingdata_send, modelv2_send, model_output, v_ego, steer_delay,
+                     publish_state, meta_main.frame_id, meta_extra.frame_id, frame_id,
                      frame_drop_ratio, meta_main.timestamp_eof, model_execution_time, live_calib_seen)
 
       desire_state = modelv2_send.modelV2.meta.desireState

selfdrive/modeld/models/commonmodel.cc

@@ -1,36 +1,30 @@
 #include "selfdrive/modeld/models/commonmodel.h"
 
-#include <cassert>
 #include <cmath>
 #include <cstring>
 
 #include "common/clutil.h"
 
-ModelFrame::ModelFrame(cl_device_id device_id, cl_context context) {
+DrivingModelFrame::DrivingModelFrame(cl_device_id device_id, cl_context context) : ModelFrame(device_id, context) {
   input_frames = std::make_unique<uint8_t[]>(buf_size);
-
-  q = CL_CHECK_ERR(clCreateCommandQueue(context, device_id, 0, &err));
-  y_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, MODEL_WIDTH * MODEL_HEIGHT, NULL, &err));
-  u_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (MODEL_WIDTH / 2) * (MODEL_HEIGHT / 2), NULL, &err));
-  v_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (MODEL_WIDTH / 2) * (MODEL_HEIGHT / 2), NULL, &err));
+  //input_frames_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, buf_size, NULL, &err));
   img_buffer_20hz_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, 5*frame_size_bytes, NULL, &err));
   region.origin = 4 * frame_size_bytes;
   region.size = frame_size_bytes;
   last_img_cl = CL_CHECK_ERR(clCreateSubBuffer(img_buffer_20hz_cl, CL_MEM_READ_WRITE, CL_BUFFER_CREATE_TYPE_REGION, &region, &err));
 
-  transform_init(&transform, context, device_id);
   loadyuv_init(&loadyuv, context, device_id, MODEL_WIDTH, MODEL_HEIGHT);
+  init_transform(device_id, context, MODEL_WIDTH, MODEL_HEIGHT);
 }
 
-uint8_t* ModelFrame::prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3 &projection, cl_mem *output) {
-  transform_queue(&this->transform, q,
-                yuv_cl, frame_width, frame_height, frame_stride, frame_uv_offset,
-                y_cl, u_cl, v_cl, MODEL_WIDTH, MODEL_HEIGHT, projection);
+uint8_t* DrivingModelFrame::prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3& projection, cl_mem* output) {
+  run_transform(yuv_cl, MODEL_WIDTH, MODEL_HEIGHT, frame_width, frame_height, frame_stride, frame_uv_offset, projection);
 
   for (int i = 0; i < 4; i++) {
     CL_CHECK(clEnqueueCopyBuffer(q, img_buffer_20hz_cl, img_buffer_20hz_cl, (i+1)*frame_size_bytes, i*frame_size_bytes, frame_size_bytes, 0, nullptr, nullptr));
   }
   loadyuv_queue(&loadyuv, q, y_cl, u_cl, v_cl, last_img_cl);
+
   if (output == NULL) {
     CL_CHECK(clEnqueueReadBuffer(q, img_buffer_20hz_cl, CL_TRUE, 0, frame_size_bytes, &input_frames[0], 0, nullptr, nullptr));
     CL_CHECK(clEnqueueReadBuffer(q, last_img_cl, CL_TRUE, 0, frame_size_bytes, &input_frames[MODEL_FRAME_SIZE], 0, nullptr, nullptr));
@@ -46,13 +40,30 @@ uint8_t* ModelFrame::prepare(cl_mem yuv_cl, int frame_width, int frame_height, i
   }
 }
 
-ModelFrame::~ModelFrame() {
-  transform_destroy(&transform);
+DrivingModelFrame::~DrivingModelFrame() {
+  deinit_transform();
   loadyuv_destroy(&loadyuv);
   CL_CHECK(clReleaseMemObject(img_buffer_20hz_cl));
   CL_CHECK(clReleaseMemObject(last_img_cl));
-  CL_CHECK(clReleaseMemObject(v_cl));
-  CL_CHECK(clReleaseMemObject(u_cl));
-  CL_CHECK(clReleaseMemObject(y_cl));
   CL_CHECK(clReleaseCommandQueue(q));
-}
+}
+
+
+MonitoringModelFrame::MonitoringModelFrame(cl_device_id device_id, cl_context context) : ModelFrame(device_id, context) {
+  input_frames = std::make_unique<uint8_t[]>(buf_size);
+  //input_frame_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, buf_size, NULL, &err));
+
+  init_transform(device_id, context, MODEL_WIDTH, MODEL_HEIGHT);
+}
+uint8_t* MonitoringModelFrame::prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3& projection, cl_mem* output) {
+  run_transform(yuv_cl, MODEL_WIDTH, MODEL_HEIGHT, frame_width, frame_height, frame_stride, frame_uv_offset, projection);
+  CL_CHECK(clEnqueueReadBuffer(q, y_cl, CL_TRUE, 0, MODEL_FRAME_SIZE * sizeof(uint8_t), input_frames.get(), 0, nullptr, nullptr));
+  clFinish(q);
+  //return &y_cl;
+  return input_frames.get();
+}
+
+MonitoringModelFrame::~MonitoringModelFrame() {
+  deinit_transform();
+  CL_CHECK(clReleaseCommandQueue(q));
+}

selfdrive/modeld/models/commonmodel.h

@@ -2,6 +2,7 @@
 
 #include <cfloat>
 #include <cstdlib>
+#include <cassert>
 
 #include <memory>
 
@@ -18,9 +19,56 @@
 
 class ModelFrame {
 public:
-  ModelFrame(cl_device_id device_id, cl_context context);
-  ~ModelFrame();
-  uint8_t* prepare(cl_mem yuv_cl, int width, int height, int frame_stride, int frame_uv_offset, const mat3& transform, cl_mem *output);
+  ModelFrame(cl_device_id device_id, cl_context context) {
+    q = CL_CHECK_ERR(clCreateCommandQueue(context, device_id, 0, &err));
+  }
+  virtual ~ModelFrame() {}
+  virtual uint8_t* prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3& projection, cl_mem* output) { return NULL; }
+  /*
+  uint8_t* buffer_from_cl(cl_mem *in_frames, int buffer_size) {
+    CL_CHECK(clEnqueueReadBuffer(q, *in_frames, CL_TRUE, 0, buffer_size, input_frames.get(), 0, nullptr, nullptr));
+    clFinish(q);
+    return &input_frames[0];
+  }
+  */
+
+  int MODEL_WIDTH;
+  int MODEL_HEIGHT;
+  int MODEL_FRAME_SIZE;
+  int buf_size;
+
+protected:
+  cl_mem y_cl, u_cl, v_cl;
+  Transform transform;
+  cl_command_queue q;
+  std::unique_ptr<uint8_t[]> input_frames;
+
+  void init_transform(cl_device_id device_id, cl_context context, int model_width, int model_height) {
+    y_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, model_width * model_height, NULL, &err));
+    u_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (model_width / 2) * (model_height / 2), NULL, &err));
+    v_cl = CL_CHECK_ERR(clCreateBuffer(context, CL_MEM_READ_WRITE, (model_width / 2) * (model_height / 2), NULL, &err));
+    transform_init(&transform, context, device_id);
+  }
+
+  void deinit_transform() {
+    transform_destroy(&transform);
+    CL_CHECK(clReleaseMemObject(v_cl));
+    CL_CHECK(clReleaseMemObject(u_cl));
+    CL_CHECK(clReleaseMemObject(y_cl));
+  }
+
+  void run_transform(cl_mem yuv_cl, int model_width, int model_height, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3& projection) {
+    transform_queue(&transform, q,
+        yuv_cl, frame_width, frame_height, frame_stride, frame_uv_offset,
+        y_cl, u_cl, v_cl, model_width, model_height, projection);
+  }
+};
+
+class DrivingModelFrame : public ModelFrame {
+public:
+  DrivingModelFrame(cl_device_id device_id, cl_context context);
+  ~DrivingModelFrame();
+  uint8_t* prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3& projection, cl_mem* output);
 
   const int MODEL_WIDTH = 512;
   const int MODEL_HEIGHT = 256;
@@ -29,10 +77,22 @@ public:
   const size_t frame_size_bytes = MODEL_FRAME_SIZE * sizeof(uint8_t);
 
 private:
-  Transform transform;
   LoadYUVState loadyuv;
-  cl_command_queue q;
-  cl_mem y_cl, u_cl, v_cl, img_buffer_20hz_cl, last_img_cl;
+  cl_mem img_buffer_20hz_cl, last_img_cl;//, input_frames_cl;
   cl_buffer_region region;
-  std::unique_ptr<uint8_t[]> input_frames;
-};
+};
+
+class MonitoringModelFrame : public ModelFrame {
+public:
+  MonitoringModelFrame(cl_device_id device_id, cl_context context);
+  ~MonitoringModelFrame();
+  uint8_t* prepare(cl_mem yuv_cl, int frame_width, int frame_height, int frame_stride, int frame_uv_offset, const mat3& projection, cl_mem* output);
+
+  const int MODEL_WIDTH = 1440;
+  const int MODEL_HEIGHT = 960;
+  const int MODEL_FRAME_SIZE = MODEL_WIDTH * MODEL_HEIGHT;
+  const int buf_size = MODEL_FRAME_SIZE;
+
+private:
+  // cl_mem input_frame_cl;
+};

selfdrive/modeld/models/commonmodel.pxd

@@ -14,5 +14,13 @@ cdef extern from "common/clutil.h":
 cdef extern from "selfdrive/modeld/models/commonmodel.h":
   cppclass ModelFrame:
     int buf_size
-    ModelFrame(cl_device_id, cl_context)
+    # unsigned char * buffer_from_cl(cl_mem*, int);
     unsigned char * prepare(cl_mem, int, int, int, int, mat3, cl_mem*)
+
+  cppclass DrivingModelFrame:
+    int buf_size
+    DrivingModelFrame(cl_device_id, cl_context)
+
+  cppclass MonitoringModelFrame:
+    int buf_size
+    MonitoringModelFrame(cl_device_id, cl_context)

selfdrive/modeld/models/commonmodel_pyx.pyx

@@ -4,11 +4,12 @@
 import numpy as np
 cimport numpy as cnp
 from libc.string cimport memcpy
+from libc.stdint cimport uintptr_t
 
 from msgq.visionipc.visionipc cimport cl_mem
 from msgq.visionipc.visionipc_pyx cimport VisionBuf, CLContext as BaseCLContext
 from .commonmodel cimport CL_DEVICE_TYPE_DEFAULT, cl_get_device_id, cl_create_context
-from .commonmodel cimport mat3, ModelFrame as cppModelFrame
+from .commonmodel cimport mat3, ModelFrame as cppModelFrame, DrivingModelFrame as cppDrivingModelFrame, MonitoringModelFrame as cppMonitoringModelFrame
 
 
 cdef class CLContext(BaseCLContext):
@@ -23,11 +24,17 @@ cdef class CLMem:
     mem.mem = <cl_mem*> cmem
     return mem
 
+  @property
+  def mem_address(self):
+    return <uintptr_t>(self.mem)
+
+def cl_from_visionbuf(VisionBuf buf):
+  return CLMem.create(<void*>&buf.buf.buf_cl)
+
+
 cdef class ModelFrame:
   cdef cppModelFrame * frame
-
-  def __cinit__(self, CLContext context):
-    self.frame = new cppModelFrame(context.device_id, context.context)
+  cdef int buf_size
 
   def __dealloc__(self):
     del self.frame
@@ -42,4 +49,28 @@ cdef class ModelFrame:
       data = self.frame.prepare(buf.buf.buf_cl, buf.width, buf.height, buf.stride, buf.uv_offset, cprojection, output.mem)
     if not data:
       return None
-    return np.asarray(<cnp.uint8_t[:self.frame.buf_size]> data)
+
+    return np.asarray(<cnp.uint8_t[:self.buf_size]> data)
+    # return CLMem.create(data)
+
+  # def buffer_from_cl(self, CLMem in_frames):
+  #   cdef unsigned char * data2
+  #   data2 = self.frame.buffer_from_cl(in_frames.mem, self.buf_size)
+  #   return np.asarray(<cnp.uint8_t[:self.buf_size]> data2)
+
+
+cdef class DrivingModelFrame(ModelFrame):
+  cdef cppDrivingModelFrame * _frame
+
+  def __cinit__(self, CLContext context):
+    self._frame = new cppDrivingModelFrame(context.device_id, context.context)
+    self.frame = <cppModelFrame*>(self._frame)
+    self.buf_size = self._frame.buf_size
+
+cdef class MonitoringModelFrame(ModelFrame):
+  cdef cppMonitoringModelFrame * _frame
+
+  def __cinit__(self, CLContext context):
+    self._frame = new cppMonitoringModelFrame(context.device_id, context.context)
+    self.frame = <cppModelFrame*>(self._frame)
+    self.buf_size = self._frame.buf_size

selfdrive/modeld/parse_model_outputs.py

@@ -96,8 +96,6 @@ class Parser:
                    out_shape=(ModelConstants.LEAD_TRAJ_LEN,ModelConstants.LEAD_WIDTH))
     if 'lat_planner_solution' in outs:
       self.parse_mdn('lat_planner_solution', outs, in_N=0, out_N=0, out_shape=(ModelConstants.IDX_N,ModelConstants.LAT_PLANNER_SOLUTION_WIDTH))
-    if 'desired_curvature' in outs:
-      self.parse_mdn('desired_curvature', outs, in_N=0, out_N=0, out_shape=(ModelConstants.DESIRED_CURV_WIDTH,))
     for k in ['lead_prob', 'lane_lines_prob', 'meta']:
       self.parse_binary_crossentropy(k, outs)
     self.parse_categorical_crossentropy('desire_state', outs, out_shape=(ModelConstants.DESIRE_PRED_WIDTH,))

selfdrive/modeld/runners/onnxmodel.py

@@ -1,39 +1,12 @@
-import onnx
-import itertools
 import os
+import onnx
 import sys
 import numpy as np
 from typing import Any
 
 from openpilot.selfdrive.modeld.runners.runmodel_pyx import RunModel
+from openpilot.selfdrive.modeld.runners.ort_helpers import convert_fp16_to_fp32, ORT_TYPES_TO_NP_TYPES
 
-ORT_TYPES_TO_NP_TYPES = {'tensor(float16)': np.float16, 'tensor(float)': np.float32, 'tensor(uint8)': np.uint8}
-
-def attributeproto_fp16_to_fp32(attr):
-  float32_list = np.frombuffer(attr.raw_data, dtype=np.float16)
-  attr.data_type = 1
-  attr.raw_data = float32_list.astype(np.float32).tobytes()
-
-def convert_fp16_to_fp32(onnx_path_or_bytes):
-  if isinstance(onnx_path_or_bytes, bytes):
-    model = onnx.load_from_string(onnx_path_or_bytes)
-  elif isinstance(onnx_path_or_bytes, str):
-    model = onnx.load(onnx_path_or_bytes)
-
-  for i in model.graph.initializer:
-    if i.data_type == 10:
-      attributeproto_fp16_to_fp32(i)
-  for i in itertools.chain(model.graph.input, model.graph.output):
-    if i.type.tensor_type.elem_type == 10:
-      i.type.tensor_type.elem_type = 1
-  for i in model.graph.node:
-    if i.op_type == 'Cast' and i.attribute[0].i == 10:
-      i.attribute[0].i = 1
-    for a in i.attribute:
-      if hasattr(a, 't'):
-        if a.t.data_type == 10:
-          attributeproto_fp16_to_fp32(a.t)
-  return model.SerializeToString()
 
 def create_ort_session(path, fp16_to_fp32):
   os.environ["OMP_NUM_THREADS"] = "4"
@@ -49,14 +22,14 @@ def create_ort_session(path, fp16_to_fp32):
     provider = 'OpenVINOExecutionProvider'
   elif 'CUDAExecutionProvider' in ort.get_available_providers() and 'ONNXCPU' not in os.environ:
     options.intra_op_num_threads = 2
-    provider = ('CUDAExecutionProvider', {'cudnn_conv_algo_search': 'DEFAULT'})
+    provider = ('CUDAExecutionProvider', {'cudnn_conv_algo_search': 'EXHAUSTIVE'})
   else:
     options.intra_op_num_threads = 2
     options.execution_mode = ort.ExecutionMode.ORT_SEQUENTIAL
     options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
     provider = 'CPUExecutionProvider'
 
-  model_data = convert_fp16_to_fp32(path) if fp16_to_fp32 else path
+  model_data = convert_fp16_to_fp32(onnx.load(path)) if fp16_to_fp32 else path
   print("Onnx selected provider: ", [provider], file=sys.stderr)
   ort_session = ort.InferenceSession(model_data, options, providers=[provider])
   print("Onnx using ", ort_session.get_providers(), file=sys.stderr)

selfdrive/modeld/runners/ort_helpers.py

@@ -0,0 +1,36 @@
+import onnx
+import onnxruntime as ort
+import numpy as np
+import itertools
+
+ORT_TYPES_TO_NP_TYPES = {'tensor(float16)': np.float16, 'tensor(float)': np.float32, 'tensor(uint8)': np.uint8}
+
+def attributeproto_fp16_to_fp32(attr):
+  float32_list = np.frombuffer(attr.raw_data, dtype=np.float16)
+  attr.data_type = 1
+  attr.raw_data = float32_list.astype(np.float32).tobytes()
+
+def convert_fp16_to_fp32(model):
+  for i in model.graph.initializer:
+    if i.data_type == 10:
+      attributeproto_fp16_to_fp32(i)
+  for i in itertools.chain(model.graph.input, model.graph.output):
+    if i.type.tensor_type.elem_type == 10:
+      i.type.tensor_type.elem_type = 1
+  for i in model.graph.node:
+    if i.op_type == 'Cast' and i.attribute[0].i == 10:
+      i.attribute[0].i = 1
+    for a in i.attribute:
+      if hasattr(a, 't'):
+        if a.t.data_type == 10:
+          attributeproto_fp16_to_fp32(a.t)
+  return model.SerializeToString()
+
+
+def make_onnx_cpu_runner(model_path):
+  options = ort.SessionOptions()
+  options.intra_op_num_threads = 4
+  options.execution_mode = ort.ExecutionMode.ORT_SEQUENTIAL
+  options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
+  model_data = convert_fp16_to_fp32(onnx.load(model_path))
+  return ort.InferenceSession(model_data,  options, providers=['CPUExecutionProvider'])

selfdrive/pandad/pandad.cc

@@ -432,6 +432,7 @@ void process_peripheral_state(Panda *panda, PubMaster *pm, bool no_fan_control)
 
     if (ir_pwr != prev_ir_pwr || sm.frame % 100 == 0 || ir_pwr >= 50.0) {
       panda->set_ir_pwr(ir_pwr);
+      Hardware::set_ir_power(ir_pwr);
       prev_ir_pwr = ir_pwr;
     }
   }