Slides - AstaZero
Transcription
Slides - AstaZero
RADAR REFLECTORS AND CLOUD DATA – AN ENABLER OF ROBUST VEHICLE AUTOMATION? Alexey Voronov, Johan Wedlin, Cristofer Englund, Viktoria Swedish ICT Johan Hultén, Sentient+, [email protected] Follow us @ViktoriaSweICT INTRODUCTION • Vehicle positioning is important for traffic safety, maintenance and autonomous driving • Painted lane markings are not enough: they wear off, and are not visible in snow and heavy rain • GPS signal is not sufficient due to limited coverage, accuracy and scrambling • Landmarks-based positioning requires expensive 3D-scanned maps • Autonomous driving requires high level of redundancy in order to not depend on the driver as a backup • Sentient+ initiated a pre-study Vehicle Positioning with Radar Reflectors. The pre-study was financed by Swedish Department of Transportation (Trafikverket) and performed in collaboration with Viktoria Swedish ICT FUNCTIONS AND REQUIREMENTS Function Safety requirement Comfort requirment Lateral control: lane-keeping Keep within the lane markings. No driver as a backup à stricter requirements than for LKA (failoperational, backup) Enough information for smooth control (e.g. soften the curves) Longitudinal control: speed adjustment Keep within the speed limits, do not miss exits and position properly on intersections Enough information for smooth control (e.g. slow down before a curve) Safe Stop Find an appropriate spot without GPS or camera - TECHNOLOGY TODAY Markings Sensors Pros Cons Painted lane markings Camera Cheap and simple Paint wear; Camera can't handle snow, rain, fog, glare, backlight Snow pins (snökäppar) Camera Available, cheap, work in snow Flyttas lätt, inte lämplig på flerfilig väg, kameraproblematik Traffic sign Camera (and partially radar) Available, rich information, useful for drivers Few variants, require text recognition, camera problems QR-code (e.g. on poles) – machine-readable sign Camera Much information on a small area Human driver can't interpret QR. Camera problems. Cat's eye in the asfalt and in Camera the guardrail Cheap, visible in the darkness Poorly visible in the daylight. Cambera problems. Guardrail Added passive safety Not applicable for muliti-lane road Cheap? Not applicable for muliti-lane road Camera and radar Delineator poles (kantstolpar Camera and radar 50 m) Landmarks (curb, tree, house/corner) LIDAR, stereocamera, camera Lots of objects + Structure-From-Motion Change often, complex algorithms Laser reflector (e.g. SICK) – a special landmark LIDAR Accurate Expensive Accurate positioning+ detailed 3D map GPS (RTK Lantmäteriet) + Sensor fusion + map Lateral and longitudinal position Outdated maps, dependency on GPS/GLONAS/... Metal wire in asfalt EM sensor Simple Not flexible, new equipment Magnets Hall-effect sensor Snow plowable, do not distract the driver New equipmnet. EM noise (bridges, trams, etc) RFID RF Rich information, exist passive och active New equipment. Uncertain life span and weather resistance Radar reflectors Radar Existing equipment Low accuracy? Need for new algorithms. A POSSIBLE SOLUTION Pavement device/apparatus • Cat's eye with radar reflector • About 20 meters apart • Identifier for longitudinal positioning • A LED-light for manually-driven vehicles? • A magnet too? Absolute positioning using cloud data • Computed from approximate (GPS) vehicle position and reflector data from the cloud (positions, identification) WHAT IS A RADAR REFLECTOR Corner reflector Reflectors can be oriented differently Aluminium can bottom Guard rail poles Several snow-plowable pavement markers from USA and Canada REFLECTORS AND CAT'S EYES DISTANCE BETWEEN REFLECTORS • • Goal: stay about ±50 cm from centerline Assume known reflector positions (±10 cm) • Measure with RTK GPS at installation, store in a database (road authority or contractor) • Lane: US: 2.7-4.6 m, Autobahn: 3.75 m Assume good radar distance accuracy (±10 cm) • 10 cm is radar accuracy at ca 10 meter • short range wide angle / side radar • Driving only on the odometer and IMU (gyro +accelerometer+compass), how far can we drive in the wind and on the rough road before we lose the remaining 30 cm sideways? • Course deviation is about 1 cm/m --> max distance 25 meter 60° • Shorter distance at the curves? • Varying distance for longitudinal coding/identification? ±20 cm ±50 cm 3m Radar distance to the lane edge: IDENTIFICATION FOR LONGITUDINAL POSITIONING Safe Stop: find a parking or a roadside/shoulder (maybe known from the map) • We know roughly where we are, in the worst case up to 1km accuracy? • All we need is a unique reflector IDs (or their sequence) in this area • • • 1 km and 20 m distance -> 50 reflectors/km 50 unique ID within 1 km requires 6 bits (26 = 64) ID can be spread out into a sequence of reflectors 1. Different Radar Cross Section 2. Polarization 3. Distance-pattern • 18-20-22 meter, mixed into "unique" sequence (------|----|-----|------|----|-----|-----) • Coding theory for the best sequence (012-012-012-012 or 120-201-020-221?) • Can distract drivers if distance variations are noticeable 4. Camera-readable sign (e.g. QR-code or several cat's eyes per reflector) • Camera: 30° angle-of-view, 10 meters distance, 2000 pixels wide • 10m * tan 30° / 2000pix = 2.8 mm/pix 4 5. Several reflectors in a row • Radar resolution is a big limitation 3 ROADWORKS? • Put a rubber mat with multiple reflectors across the road in good time for the roadworks ahead • The mat marks the termination of autonomous driving. Either the driver takes over or Safe Stop is performed (area for this should be created) SNOW? Plowed snow Road Cat's eye Snökäppar Kantstolpar POTENTIAL BENEFITS • • • • • • • Important step towards allowing autonomous driving Redundancy for safety functions in autonomous driving Enhanced lane keeping Reduced lane departure accidents Allows narrower lane width Reduced guard rail repairs Reduced dependency on GPS / GLONAS (Civil Defense) NEXT STEPS ? • So far research indicates that this is a promising direction. More investigations are needed before we can say that this could work. • Evaluate with actual radars • Accuracy of distance measurement; angular resolution • Size of the reflector, Radar Cross Section • Possibility of polarization • Water-repellent and snowplowable design; combination with cat's eye/magnet/LED • Verify the minimum required distance between the reflectors • Longitudinal coding • Alt 1: Cat's eye every 5 meters, radar-reflector every other-third-forth • Alt 2: driver acceptance of varied distance (18-20-22 meter) • Best code ("the most unique" sequence, error correcting code/checksum) • Demonstrate prototype at AstaZero