Ring Road Experiment For Driving Safety Analysis Beijing, 2011 Spring POSS , Peking University Ring Road Experiment For Driving Safety Analysis Beijing, 2011 Spring
Experiment Objective Objectives Collect multi-modal sensing data in ring-road/free-way driving scenario Lasers sensing other traffic participants in vicinity and moving obstacle detection and tracking GPS data for host vehicle positioning and trajectory recording Panoramic camera for ground truth Driving behavior analysis The driving behavior analysis includes several meaning: To learn from human driver’s driving behavior in order to refine the trajectories or motion planning strategies; to investigate the effect of other traffic participants on our host vehicle so that to give risk assessment that is proper to describe the potential danger around our robot. In addition we can study the relation between human driving strategy and the vicinity traffic to learn some decision making progress in behavior layer of the motion planner, such as the switching between different strategies in different situation.(Lane changing/overtaking/car following etc.)
Sensor Setting LadyBug GPS/IMU LMS URG This photo is not taken when we did the experiment, so the sensor setting is illustrated with virtual sensors. In our experiment we only use HAYUKO URG laser(small)*3, a LMS291, GPS/IMU and the ladybug. URG
Sensing Range Coverage SICK LMS LadyBug Panoramic camera RM FM FL FR HAKUYO URG Lasers Laser Coverage Front Middle SICK LMS Front Left(Right) HAKUYO URG Rear Middle The gray circle represents the ladybug is 360°, the blue area is covered with urg and the red one is LMS. In fact the LMS at front middle is used only to obtain a further sensing range and in case the URG is not available. The URGs have already covered the proposed sensing area.
Experiment Design Experiment environment Experiment time 3rd ring road in Beijing Whole length: 96.53km Design speed: 80km/h Experiment time 2011/05/30 20:38—22:50(UTC+8) The experiment is carried at night which is not the rush hour.
Experiment Design: Beijing 3rd Ring Road
Data Example: Lasers & Panoramic Camera 4 Lasers + GPS Panoramic Camera
Data Example: GPS Data GPS failure GPS points GPS failure
Data Example: GPS Data Large GPS error
Data Processing Steps BM Laser Calibration Laser Synchronization Laser calibration Schematic Steps Laser Calibration Laser Synchronization Laser-GPS interpolation BM FM FL FR Box This slide shows the laser calibration procedure. By looking at the same box at different position around the car using multi lasers, we can calibrate the relationship between these lasers.
Data Analysis Generating Occupancy Grid Map Using GPS + laser data Including host vehicle trajectory Including static obstacles
Data Analysis: Occupancy Grid Map Generation This is a 1:12 long video to replay part of our exp on OGM, the “Lane changing” behavior is distinguished from the “overtaking” behavior since the trajectory didn’t return to it’s former line.
Data Analysis SLAM with Moving Obstacle Detection and Tracking Yellow: host vehicle trajectory Orange: moving objects and their trajectories Green: static obstacles Cyan: undecided
Data Analysis GPS Data Interpolation Why need interpolation? Yaw data from GPS is not always available To get pos\yaw\curvature between GPS data points Interpolation using cubic spline Guarantee C2 continuity trajectory (with continuous yaw and curvature) to satisfy nonholonomic constraints
Raw GPS Data with yaw error The two videos show the necessity of interpolation. There are two key frames in the upper video, while no key frame in the bottom one which just show it has improved the yaw data at the same position with the key frames in the upper video by interpolation. After interpolation
Future Work Generating trajectory set from real GPS trajectory of host vehicle Extract some features(curvature, length etc) from GPS trajectory Learning human driving behavior from real data with these features and comparing the difference between motion planning results and the real trajectory of human driving vehicle Try to find some driving safety indices by learning from the state of moving obstacles around the host vehicle
Thank You!