IQ.Lvbs/IQ.Pilot
8
7
Fork 2
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

IQ.Pilot Release Commit @ 9ba2ea7

Browse Source
This commit is contained in:
IQ.Lvbs CI [bot]
2026-08-31 10:53:00 -05:00
parent 4b46753a27
commit 67a2b4955e
22 changed files with 2011 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

214
iqpilot/selfdrive/controls/radard.py
View File

@@ -13,6 +13,7 @@ from iqpilot.common.swaglog import cloudlog
from iqpilot.common.simple_kalman import KF1D
from iqdbc.car import structs
from iqdbc.car.honda.values import HONDA_RADAR_SCAN_CAPABLE
from iqdbc.car.hyundai.values import HyundaiFlags, HyundaiFlagsIQ
from iqpilot.selfdrive.controls.lib.custom_stop_distance import CustomStopDistance
@@ -29,6 +30,19 @@ V_EGO_STATIONARY = 4. # no stationary object flag below this speed
RADAR_TO_CENTER = 2.7 # (deprecated) RADAR is ~ 2.7m ahead from center of car
RADAR_TO_CAMERA = 1.52 # RADAR is ~ 1.5m ahead from center of mesh frame
# Honda radar object scan: 15Hz sweeps consumed at the 20Hz model rate, so measurement absorption is
# gated on fresh sweep data and lead selection carries continuity/staleness evidence
SCAN_SWEEP_DT = 1.0 / 15
SCAN_LEAD_PROB = 0.35
SCAN_LEAD_MIN_CYCLES = 3
SCAN_CHALLENGER_STALE_CYCLES = 2
SCAN_DISTANCE_STALE_CYCLES = 3
SCAN_DISTANCE_STALE_M = 25.0
def uses_scan_radar(CP) -> bool:
return CP.brand == "honda" and CP.carFingerprint in HONDA_RADAR_SCAN_CAPABLE and not CP.radarUnavailable
class KalmanParams:
def __init__(self, dt: float):
@@ -62,7 +76,8 @@ class Track:
self.K_K = kalman_params.K
self.kf = KF1D([[v_lead], [0.0]], self.K_A, self.K_C, self.K_K)
def update(self, d_rel: float, y_rel: float, v_rel: float, v_lead: float, measured: float):
def update(self, d_rel: float, y_rel: float, v_rel: float, v_lead: float, measured: float,
absorb_measurement: bool = True):
# relative values, copy
self.dRel = d_rel # LONG_DIST
self.yRel = y_rel # -LAT_DIST
@@ -70,18 +85,19 @@ class Track:
self.vLead = v_lead
self.measured = measured # measured or estimate
# computed velocity and accelerations
if self.cnt > 0:
# a repeated scan payload between 15Hz sweeps must not be absorbed as a second measurement
if absorb_measurement and self.cnt > 0:
self.kf.update(self.vLead)
self.vLeadK = float(self.kf.x[SPEED][0])
self.aLeadK = float(self.kf.x[ACCEL][0])
# Learn if constant acceleration
if abs(self.aLeadK) < 0.5:
self.aLeadTau.x = _LEAD_ACCEL_TAU
else:
self.aLeadTau.update(0.0)
if absorb_measurement:
# Learn if constant acceleration
if abs(self.aLeadK) < 0.5:
self.aLeadTau.x = _LEAD_ACCEL_TAU
else:
self.aLeadTau.update(0.0)
self.cnt += 1
@@ -119,18 +135,35 @@ def laplacian_pdf(x: float, mu: float, b: float):
return math.exp(-abs(x-mu)/b)
def model_association_score(track: Track, lead: capnp._DynamicStructReader, v_ego: float) -> float:
offset_vision_dist = lead.x[0] - RADAR_TO_CAMERA
prob_d = laplacian_pdf(track.dRel, offset_vision_dist, lead.xStd[0])
prob_y = laplacian_pdf(track.yRel, -lead.y[0], lead.yStd[0])
prob_v = laplacian_pdf(track.vRel + v_ego, lead.v[0], lead.vStd[0])
# This isn't exactly right, but it's a good heuristic
return prob_d * prob_y * prob_v
def track_agrees_with_model(track: Track, lead: capnp._DynamicStructReader, v_ego: float, strict: bool) -> bool:
dist_scale, dist_floor, vel_limit, y_std_scale, y_floor = \
(0.25, 5.0, 10.0, 1.0, 1.0) if strict else (0.40, 8.0, 13.0, 2.0, 1.5)
vision_dist = lead.x[0] - RADAR_TO_CAMERA
dist_ok = abs(track.dRel - vision_dist) < max(abs(vision_dist) * dist_scale, dist_floor)
vel_ok = (abs(track.vRel + v_ego - lead.v[0]) < vel_limit) or (v_ego + track.vRel > 3)
lat_ok = abs(track.yRel + lead.y[0]) < max(y_floor, y_std_scale * max(float(lead.yStd[0]), 0.2))
return dist_ok and vel_ok and lat_ok
def scan_low_speed_candidate(track: Track, v_ego: float) -> bool:
# require a few real cycles before a radar-only low-speed takeover
return track.cnt >= SCAN_LEAD_MIN_CYCLES and track.potential_low_speed_lead(v_ego)
def match_vision_to_track(v_ego: float, lead: capnp._DynamicStructReader, tracks: dict[int, Track]):
offset_vision_dist = lead.x[0] - RADAR_TO_CAMERA
def prob(c):
prob_d = laplacian_pdf(c.dRel, offset_vision_dist, lead.xStd[0])
prob_y = laplacian_pdf(c.yRel, -lead.y[0], lead.yStd[0])
prob_v = laplacian_pdf(c.vRel + v_ego, lead.v[0], lead.vStd[0])
# This isn't exactly right, but it's a good heuristic
return prob_d * prob_y * prob_v
track = max(tracks.values(), key=prob)
track = max(tracks.values(), key=lambda c: model_association_score(c, lead, v_ego))
# if no 'sane' match is found return -1
# stationary radar points can be false positives
@@ -161,22 +194,60 @@ def get_RadarState_from_vision(lead_msg: capnp._DynamicStructReader, v_ego: floa
def get_lead(v_ego: float, ready: bool, tracks: dict[int, Track], lead_msg: capnp._DynamicStructReader,
model_v_ego: float, CP: structs.CarParams, CP_IQ: structs.IQCarParams, low_speed_override: bool = True) -> dict[str, Any]:
model_v_ego: float, CP: structs.CarParams, CP_IQ: structs.IQCarParams, low_speed_override: bool = True,
scan_radar: bool = False, filtered_prob: float | None = None, held_track_id: int = -1) -> dict[str, Any]:
lead_prob = float(lead_msg.prob if filtered_prob is None else filtered_prob)
prob_threshold = SCAN_LEAD_PROB if scan_radar else .5
# Determine leads, this is where the essential logic happens
if len(tracks) > 0 and ready and lead_msg.prob > .5:
if len(tracks) > 0 and ready and lead_prob > prob_threshold:
track = match_vision_to_track(v_ego, lead_msg, tracks)
else:
track = None
lead_dict = {'status': False}
if track is not None:
lead_dict = track.get_RadarState(lead_msg.prob)
lead_dict = track.get_RadarState(lead_prob)
lead_dict = get_custom_yrel(CP, CP_IQ, lead_dict, lead_msg)
elif (track is None) and ready and (lead_msg.prob > .5):
elif (track is None) and ready and (lead_prob > prob_threshold):
lead_dict = get_RadarState_from_vision(lead_msg, v_ego, model_v_ego)
if low_speed_override:
low_speed_tracks = [c for c in tracks.values() if c.potential_low_speed_lead(v_ego)]
if scan_radar:
low_speed_tracks = [c for c in tracks.values() if scan_low_speed_candidate(c, v_ego)]
else:
low_speed_tracks = [c for c in tracks.values() if c.potential_low_speed_lead(v_ego)]
model_lead_available = ready and lead_prob > prob_threshold
if scan_radar:
# Keep the held radar lead through ordinary model-probability fluctuations while it stays
# coherent. With a valid model lead it must still agree with it; without one, a mature radar
# track remains eligible for continuity
held = tracks.get(held_track_id)
if held is not None and scan_low_speed_candidate(held, v_ego):
held_matches_model = (not model_lead_available or
track_agrees_with_model(held, lead_msg, v_ego, strict=True))
held_is_current = (not lead_dict.get('status', False) or
lead_dict.get('radarTrackId', -1) == held_track_id or
(lead_dict.get('status', False) and not lead_dict.get('radar', False)))
if held_is_current and held_matches_model:
lead_dict = held.get_RadarState(lead_prob)
def candidate_established(candidate: Track) -> bool:
if candidate.cnt < SCAN_LEAD_MIN_CYCLES:
return False
if not lead_dict.get('status', False):
# a mature centered scan point may provide the radar-only low-speed lead
return True
if lead_dict.get('radarTrackId', -1) == candidate.identifier:
return True
# never replace an established lead with an unrelated closer point without model evidence
# to arbitrate them
return model_lead_available and track_agrees_with_model(candidate, lead_msg, v_ego, strict=True)
low_speed_tracks = [c for c in low_speed_tracks if candidate_established(c)]
if len(low_speed_tracks) > 0:
closest_track = min(low_speed_tracks, key=lambda c: c.dRel)
@@ -204,7 +275,16 @@ class RadarD:
self.current_time = 0.0
self.tracks: dict[int, Track] = {}
self.kalman_params = KalmanParams(DT_MDL)
self.scan_radar = uses_scan_radar(CP)
# the lead KF absorbs scan measurements at the physical 15Hz sweep cadence; lead probability
# filtering stays on model-loop timing
self.kalman_params = KalmanParams(SCAN_SWEEP_DT if self.scan_radar else DT_MDL)
self.lead_prob_filters = [FirstOrderFilter(0.0, 0.2, DT_MDL) for _ in range(2)]
self.held_lead_ids = [-1, -1]
self._held_evidence_ids = [-1, -1]
self._challenger_stale_counts = [0, 0]
self._distance_stale_counts = [0, 0]
self._last_tracks_frame = -1
self.v_ego = 0.0
self.v_ego_hist = deque([0.0], maxlen=int(round(delay / DT_MDL))+1)
@@ -217,6 +297,53 @@ class RadarD:
self.custom_stop_distance = CustomStopDistance()
def _refresh_held_lead_evidence(self, lead_index: int, lead: capnp._DynamicStructReader,
lead_prob: float) -> None:
held_id = self.held_lead_ids[lead_index]
if self._held_evidence_ids[lead_index] != held_id:
self._reset_held_evidence(lead_index, held_id)
held = self.tracks.get(held_id)
if held_id < 0 or held is None or not self.ready or lead_prob <= SCAN_LEAD_PROB:
self._reset_held_evidence(lead_index, held_id)
return
strict_match = track_agrees_with_model(held, lead, self.v_ego, strict=True)
relaxed_match = track_agrees_with_model(held, lead, self.v_ego, strict=False)
# evidence arm 1: another live track scores better against the model while the held one no
# longer passes even relaxed continuity. Releasing the hold never selects that challenger; the
# strict-match path in get_lead stays the only way it becomes the radar lead
if relaxed_match:
self._challenger_stale_counts[lead_index] = 0
else:
best = max(self.tracks.values(), key=lambda c: model_association_score(c, lead, self.v_ego))
if best.identifier != held_id and \
model_association_score(best, lead, self.v_ego) > model_association_score(held, lead, self.v_ego):
self._challenger_stale_counts[lead_index] += 1
else:
self._challenger_stale_counts[lead_index] = 0
# evidence arm 2: gross absolute range disagreement, with a strict match staying authoritative
# even when model uncertainty would permit the error
distance_mismatch = abs(held.dRel - (lead.x[0] - RADAR_TO_CAMERA))
if strict_match:
self._distance_stale_counts[lead_index] = 0
elif distance_mismatch > SCAN_DISTANCE_STALE_M:
self._distance_stale_counts[lead_index] += 1
else:
self._distance_stale_counts[lead_index] = 0
if (self._challenger_stale_counts[lead_index] >= SCAN_CHALLENGER_STALE_CYCLES or
self._distance_stale_counts[lead_index] >= SCAN_DISTANCE_STALE_CYCLES):
self.held_lead_ids[lead_index] = -1
self._reset_held_evidence(lead_index)
def _reset_held_evidence(self, lead_index: int, held_id: int = -1) -> None:
self._held_evidence_ids[lead_index] = held_id
self._challenger_stale_counts[lead_index] = 0
self._distance_stale_counts[lead_index] = 0
def update(self, sm: messaging.SubMaster, rr: car.RadarData):
self.ready = sm.seen['modelV2']
self.current_time = 1e-9*max(sm.logMonoTime.values())
@@ -227,6 +354,11 @@ class RadarD:
self.v_ego_hist.append(self.v_ego)
self.last_v_ego_frame = sm.recv_frame['carState']
sweep_fresh = True
if self.scan_radar:
sweep_fresh = sm.recv_frame['radarTracks'] != self._last_tracks_frame
self._last_tracks_frame = sm.recv_frame['radarTracks']
ar_pts = {pt.trackId: [pt.dRel, pt.yRel, pt.vRel, pt.measured] for pt in rr.points}
# *** remove missing points from meta data ***
@@ -244,7 +376,11 @@ class RadarD:
# create the track if it doesn't exist or it's a new track
if ids not in self.tracks:
self.tracks[ids] = Track(ids, v_lead, self.kalman_params)
self.tracks[ids].update(rpt[0], rpt[1], rpt[2], v_lead, rpt[3])
if self.scan_radar:
measured = bool(rpt[3] and sweep_fresh)
self.tracks[ids].update(rpt[0], rpt[1], rpt[2], v_lead, measured, absorb_measurement=measured)
else:
self.tracks[ids].update(rpt[0], rpt[1], rpt[2], v_lead, rpt[3])
# *** publish radarState ***
self.radar_state_valid = sm.all_checks()
@@ -259,8 +395,34 @@ class RadarD:
model_v_ego = self.v_ego
leads_v3 = sm['modelV2'].leadsV3
if len(leads_v3) > 1:
lead_one = get_lead(self.v_ego, self.ready, self.tracks, leads_v3[0], model_v_ego, self.CP, self.CP_IQ, low_speed_override=True)
lead_two = get_lead(self.v_ego, self.ready, self.tracks, leads_v3[1], model_v_ego, self.CP, self.CP_IQ, low_speed_override=False)
if self.scan_radar:
for i in range(2):
lead_prob = float(leads_v3[i].prob)
# probability rises instantly, decays filtered: a one-cycle model dip must not drop the lead
if lead_prob > self.lead_prob_filters[i].x:
self.lead_prob_filters[i].x = lead_prob
else:
self.lead_prob_filters[i].update(lead_prob)
self._refresh_held_lead_evidence(i, leads_v3[i], self.lead_prob_filters[i].x)
lead_one = get_lead(self.v_ego, self.ready, self.tracks, leads_v3[0], model_v_ego, self.CP, self.CP_IQ,
low_speed_override=True, scan_radar=True, filtered_prob=self.lead_prob_filters[0].x,
held_track_id=self.held_lead_ids[0])
lead_two = get_lead(self.v_ego, self.ready, self.tracks, leads_v3[1], model_v_ego, self.CP, self.CP_IQ,
low_speed_override=False, scan_radar=True, filtered_prob=self.lead_prob_filters[1].x,
held_track_id=self.held_lead_ids[1])
for i, lead in enumerate((lead_one, lead_two)):
if lead.get('status', False) and lead.get('radar', False):
track_id = int(lead.get('radarTrackId', -1))
if track_id != self.held_lead_ids[i]:
self._reset_held_evidence(i, track_id)
self.held_lead_ids[i] = track_id
elif (not lead.get('status', False)) or (self.held_lead_ids[i] not in self.tracks):
self.held_lead_ids[i] = -1
else:
lead_one = get_lead(self.v_ego, self.ready, self.tracks, leads_v3[0], model_v_ego, self.CP, self.CP_IQ, low_speed_override=True)
lead_two = get_lead(self.v_ego, self.ready, self.tracks, leads_v3[1], model_v_ego, self.CP, self.CP_IQ, low_speed_override=False)
self.radar_state.leadOne = self.custom_stop_distance.apply_lead(lead_one)
self.radar_state.leadTwo = self.custom_stop_distance.apply_lead(lead_two)