IQ.Pilot Release Commit @ 0798119

iqdbc_repo/iqdbc/car/byd/values.py

@@ -0,0 +1,143 @@
+from dataclasses import dataclass, field
+from enum import IntFlag, StrEnum
+
+from iqdbc.car import ACCELERATION_DUE_TO_GRAVITY, Bus, CarSpecs, DbcDict, PlatformConfig, Platforms, structs
+from iqdbc.car.lateral import AngleSteeringLimits, AVERAGE_ROAD_ROLL, ISO_LATERAL_ACCEL
+from iqdbc.car.docs_definitions import CarDocs, CarHarness, CarParts
+from iqdbc.car.fw_query_definitions import FwQueryConfig, Request, StdQueries
+from iqdbc.car.vin import Vin
+
+Ecu = structs.CarParams.Ecu
+
+
+class CarControllerParams:
+  STEER_STEP = 2  # 50 Hz
+
+  ANGLE_LIMITS: AngleSteeringLimits = AngleSteeringLimits(
+    390,  # deg
+    ([], []),
+    ([], []),
+
+    MAX_LATERAL_ACCEL=ISO_LATERAL_ACCEL + (ACCELERATION_DUE_TO_GRAVITY * AVERAGE_ROAD_ROLL),
+    MAX_LATERAL_JERK=3.0 + (ACCELERATION_DUE_TO_GRAVITY * AVERAGE_ROAD_ROLL),
+
+    # deg/20ms. EPS faults at 12 at a standstill; 5 caused command spikes, 2 under-tracked
+    # sharp curves. Stock Veoneer max is 4.8.
+    MAX_ANGLE_RATE=3,
+  )
+
+  # Low-speed taper on the angle rate, in deg per STEER_STEP frame.
+  #
+  # The vehicle-model jerk limit scales as 1/v^2, so below a few m/s it stops binding and only
+  # the flat MAX_ANGLE_RATE is left. The lateral planner is ill-conditioned at a standstill and
+  # oscillates, and slewing the command at the full rate while the wheel is not moving walks the
+  # EPS straight from state 9 to a latched 11. Measured 2026-08-05 at 0.29 m/s: the command swung
+  # -5.9 to +2.4 deg in 220 ms against a stationary wheel and the EPS latched, taking LKAS with
+  # it. A healthy engagement at 0.9 m/s held the command within 1.1 deg of measured.
+  ANGLE_RATE_BP = [0.0, 2.0, 5.0]  # m/s
+  ANGLE_RATE_V = [0.3, 1.0, 3.0]   # deg/frame, tops out at MAX_ANGLE_RATE
+
+  # STEERING_TORQUE.DRIVER_TORQUE thresholds, derived from a drive where openpilot actually
+  # steered (route 0000000f, EPS state 10):
+  #   |torque| while openpilot steered:  p50 1.2  p90 2.7  p95 3.3  p99 5.8  max 9.8
+  #   |torque| while the human drove:    p50 0.2  p90 8.1  p95 17.2 p99 24.5 max 35.5
+  # The old 3.0 sat below what openpilot generates while steering, so it tripped its own
+  # override and dropped out within a few frames of every engage.
+  STEER_DRIVER_OVERRIDE = 12.0   # above openpilot's own max, below a deliberate grab
+
+  # Never command further than this from the actual wheel angle. The EPS latches state 11 on
+  # angle divergence, not just on a lost stream: measured 2026-08-05, the driver held the wheel
+  # at -13.4 deg while the controller wound the command out to -48.1 deg and the EPS latched at
+  # 34.7 deg of error. Normal closed-loop steering holds the error inside ~1.1 deg, so this only
+  # bites when the wheel is being physically held.
+  MAX_ANGLE_ERROR = 10.0  # deg
+
+  # comfort envelope, inside the safety cap of -3.5..+2.0
+  ACCEL_MIN = -3.0
+  ACCEL_MAX = 1.5
+
+  JERK_UP = 2.5
+  JERK_UP_LAUNCH = 4.0  # below 2 m/s, to beat the ~0.5s IPB lag off the line
+  JERK_DOWN = 5.0
+
+
+class BydSafetyFlags(IntFlag):
+  LONG_CONTROL = 1
+
+
+class BydFlags(IntFlag):
+  # The ADAS/ACC ECU is behind the relay, so its 0x32E ACC_CMD can be blocked and replaced.
+  # Set when ACC_CMD is fingerprinted on the camera-side bus.
+  GATEWAY_HARNESS = 1
+
+
+# addresses used to tell the two harness types apart
+ACC_CMD_ADDR = 0x32E
+
+
+class WMI(StrEnum):
+  BYD_AUTO = "LGX"
+
+
+class ModelYear(StrEnum):
+  R_2024 = "R"
+  S_2025 = "S"
+  T_2026 = "T"
+
+
+@dataclass
+class BydCarDocs(CarDocs):
+  package: str = "All"
+  car_parts: CarParts = field(default_factory=CarParts.common([CarHarness.custom]))
+
+
+@dataclass
+class BydPlatformConfig(PlatformConfig):
+  dbc_dict: DbcDict = field(default_factory=lambda: {Bus.pt: 'byd_sealion_7'})
+  wmis: set[WMI] = field(default_factory=set)
+  years: set[ModelYear] = field(default_factory=set)
+  vds_prefixes: set[str] = field(default_factory=set)
+
+
+class CAR(Platforms):
+  BYD_SEALION_7 = BydPlatformConfig(
+    [BydCarDocs("BYD Sealion 7 2024-25")],
+    CarSpecs(mass=2090., wheelbase=2.93, steerRatio=16.0, centerToFrontRatio=0.44),
+    wmis={WMI.BYD_AUTO},
+    years={ModelYear.R_2024, ModelYear.S_2025, ModelYear.T_2026},
+  )
+
+
+def match_fw_to_car_fuzzy(live_fw_versions, vin, offline_fw_versions) -> set[str]:
+  # VIN: LGX (WMI) + <VDS> + <year><plant><seq> (VIS). Matching on WMI + year alone would claim
+  # every BYD of that year, so a platform only matches once its VDS prefix is known.
+  vin_obj = Vin(vin)
+  year = vin_obj.vis[:1]
+
+  candidates = set()
+  for platform in CAR:
+    cfg = platform.config
+    if not cfg.vds_prefixes or vin_obj.wmi not in cfg.wmis or year not in cfg.years:
+      continue
+    if any(vin_obj.vds.startswith(p) for p in cfg.vds_prefixes):
+      candidates.add(platform)
+
+  return {str(c) for c in candidates}
+
+
+FW_QUERY_CONFIG = FwQueryConfig(
+  # BYD ECUs NRC 0xF188 (openpilot's default) but answer 0xF195
+  requests=[
+    Request(
+      [StdQueries.SUPPLIER_SOFTWARE_VERSION_REQUEST],
+      [StdQueries.SUPPLIER_SOFTWARE_VERSION_RESPONSE],
+      bus=0,
+    ),
+  ],
+  # the MPC camera answers OBD DTC scans but not the bus-0 DID sweep
+  non_essential_ecus={Ecu.fwdCamera: list(CAR)},
+  match_fw_to_car_fuzzy=match_fw_to_car_fuzzy,
+)
+
+
+DBC = CAR.create_dbc_map()