o to thong minh

Bộ mơn Ơ tơ xe Chun dụng Đại học Bách khoa Hà nội Ô TÔ THÔNG MINH INTELLIGENT VEHICLE TS ĐÀM HỒNG PHÚC 2012.4.14 　 SƠN TÂY Purpose • Investigate the motivation of adding Intelligence to a car • Explore problems and solutions • Survey the current state of research • Identify future research trends Outline • • • • • Definitions / Motivation Design Goals Problems / Solutions - Theory Current Industry Solutions Future Trend Background Active safety devices (driver assistance systems) for reducing road accidents Controller Driver Integrated Vehicle Dynamics Control Adaptive Cruise Control Vehicle Driver-Vehicle Cooperation Pre-Crash Safety System Motivation : Integrated Vehicle Technology and Robotics  Machine Intelligence : to provide service with respect to the operation task  Sensor information is used to recognize the operation modes /behaviour Driver steering behaviour adaptation ・　 lane keeping task : lane keeping control activates ・ lane changing : stability control activates Driver intends to make a lane change manoeuvre Skidding avoidance is needed to secure vehicle stability during transient steering manoeuvre Yaw moment generated by Lane keeping system If Lane-keeping system is still activated, it may cause conflict with driver steering manoeuvre Motivation and Objectives ■ 　 Conventional driver assistance systems use average characteristics of drivers 　→　 They not fit the driver preference and cause sense of discomfort 　　　→　 Accident prevention effect cannot fully be obtained as expected Accident Risk Potential High (Dangerous) Low (Safe) ■ 　 Environment adaptation ■ 　 Driver adaptation Driver state (drowsy, tired, hurry) Effective system activation Traffic environment (map, vicinity vehicles) Running Distance Driver State Good (Cautious) Bad (careless) Schematic diagram of driver assistance system Based on long-term naturalistic driving data of a driver in actual traffic condition, the following algorithms for synthesizing DA are designed (1)”Feature of usual driving behavior”, (2)“Unusual behavior detection and prediction” , (3)“Individual adaptation of driver assistance system” Driver Assistance Systems • • • • • • • • Night Vision Adaptive Cruise Control Collision Warning Collision Avoidance Driver Impairment Monitoring Advanced Driver Assistance Cooperative Infrastructure Automated Driving TTLC 　：　 Time 　 To 　 Line 　 Crossing 　 [s] D TTLC = V D V 現在の位置 TTLC large Warning TTLC 後の位置 V ： Velocity D ： Distance TTLC Small Assist Forward collision warning system brake on warni ng V Vp R Conventional collision risk index R •　 TTC (Time to collision) V − V p •　 THW (Time headway) host vehicle preceding vehicle R V Comparison between the ground truth (reference) and the estimated state Driver-Vehicle-Environment modeling based on Boosting Sequential Labeling Method Decelerating Braking Stopping Cruising Following Ground Truth Features 　 Considering the feature of state-flow 　 Low computational cost with expectable high accuracy 　 Easy to interprete the physical meanings of the model ： S1 Road sections： R2 Training ： 7-Trip data sets Estimated： trip data th 50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400 Estimated R [m] 100 R Relative distance 50 V [km/h] Vehicle velocity 0 V 50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400 Thw [s] Time headway 0 10 &[m/s] R Relative velocity Driver Ttc −1[1/s] Inverse of time to collision ax [m/s ] Longitudinal acceleration -10 -2 0.5 -0.5 Time [s] Comparisonbetween between ground truth (reference) and the estimated Comparison thethe ground truth (reference) and the estimated state state Driver ： S1 Road sections： R2 Following Braking Training ： 7-Trip data sets Decelerating Stopping Cruising Estimated ： 8th trip data Ground Truth 50 100 150 200 250 300 350 400 50 100 150 200 250 300 350 400 Estimation Behavior Accuracy [%] Following 98.2 Braking 91.7 Cruising 96.1 Decelerating 89.0 Stopping 100 Time [s] The estimation result of each driving state shows high accuracy of the recognition algorithm Driver Model D esired velocity (km /h) 60 0 350 700 Difference>45% Time (s) Driver Driver A Driver B Fuel economy Running distance km/l 3.50 5.17 km 49.3 50.9 Analysis fuel economy of each driver Driver model NAGAI LAB ＤＳ　 experimental condition Human longitudinal control driver-vehicle system in closed-loop Flat and straight road, µ=0.8 Hybrid Truck, m=5500kg 80 V [km/h] 60 40 20 0 20 40 60 80 100 120 0.8 ax [m/s2] 0.6 0.4 0.2 -0.2 20 40 60 Time [s] 80 100 120 Five drivers conducted experiments Normal Driving ： times Hurry Driving ： times Driver model NAGAI LAB Comparison of hurry and normal driving experimental results (D4) Vehicle velocity [km/h] Headway distance [m] Relative velocity [m/s] 50 0 20 Normal Driving 40 60 80 100 120 Hurry Driving 50 0 20 40 60 80 100 120 20 40 60 80 100 120 20 40 60 Time [s] 80 100 120 15 10 -5 100 Accelerator pedal stroke [%] 50 0 In the hurry driving situation both headway distance and velocity errors are smaller Human accelerator pedal operations are wavier Velocity Tracking Performance NAGAI LAB The relationship between velocity tracking performance with fuel economy Ｖ f following vehicle velocity Ｖ p Preceding vehicle velocity Ｖ The average velocity error t fn eV _ avr = t0 tfn t [s] t0 Better in velocity tracking performance will give better fuel economy Good -0.478 Bad -0.202 ( V − V ) dt (t fn − t0 ) p f ∫ The effect of velocity tracking performance on fuel economy in normal driving is stronger than that of hurry driving condition Accelerator Pedal Operation Characteristics NAGAI LAB The relationship between pedal operation characteristics with fuel economy Good Bad Large variations in the accelerator pedal operation gives larger negative effect on fuel consumption of the vehicle Accelerator pedal variance Phm_var [%] There are strong relationship between accelerator pedal operation characteristics and velocity tracking performance with fuel economy Therefore the driver model need to describe driver’s accelerator pedal operation and represent of vehicle control process 13 Driver model NAGAI LAB Driver operation HEV Vf ax Host vehicle Vp Rstop Preceding vehicle VfThw x The driver attempt to diminish distance error between actual headway distance and his desired headway distance Driver accelerator model  −1  CV & Phm =   Phm +  τ h   τh Hx τh ∆V = V p − V f ： Relative velocity ∆R = R − ThwV f： Headway distance error T HV   V p ∆R ∆V  τh  Pa ： Accelerator pedal stroke V p ： Preceding vehicle velocity CV Preceding velocity feed-forward gain V f ： Following vehicle velocity HV Relative velocity feedback gain R ： Headway distance H x Headway distance feedback gain τ h Human time delay Thw ： Time headway Driver model NAGAI LAB Identified Driver Model Parameters Hurry driving D2 D3 D4 D5 HV[s/m] 0.009 0.000 0.000 0.000 0.002 Hx [1/m] 0.009 0.019 0.024 0.022 0.005 CV[s/m] 0.030 0.032 0.033 0.031 0.032 Thw [s] 3.7 2.3 2.6 2.9 3.7 HV[s/m] 0.287 0.000 0.098 0.070 0.049 Hx [1/m] 0.028 0.025 0.029 0.040 0.007 CV[s/m] 0.033 0.031 0.033 0.033 0.031 Thw [s] 0.78 1.37 0.89 1.30 1.16 Verify the accuracy of the driver model In normal driving situation, the time headway, Thw are large and driver control gain Hx and HV are small In hurry driving situation Thw are small and driver control gain Hx and HV are large 80 60 V [km/h] Normal driving D1 40 20 0 20 40 60 80 100 120 20 40 60 Time [s] 80 100 120 0.8 0.6 ax [m/s2] 　 Driver 0.4 0.2 -0.2 15 Driver model NAGAI LAB The comparison of normal driving simulation results and experimental data (D4) Vehicle velocity [km/h] Headway distance [m] Relative velocity [m/s] Accelerator pedal stroke [%] 50 Vp 0 20 40 60 80 Expr Data 100 Sim 120 50 Expr.Data 0 Sim 20 40 60 80 100 120 20 40 60 80 100 120 20 40 60 Time [s] 80 100 120 20 10 -10 100 50 0 In normal driving situation, the headway distance, velocity errors and accelerator pedal stroke of simulation results are well-fitted with the experimental data 16 Driver model NAGAI LAB Vehicle velocity [km/h] V [km/h] The comparison of hurry driving simulation results and experimental data (D4) 50 V Vexpr p 0 20 40 60 80 Vsim 100 120 30 Headway distance [m] Hd [m] Vexpr Vsim 20 10 10 20 30 40 50 60 70 80 90 100 Relative velocity [m/s] Vr [m/s] 10 -5 20 40 60 80 100 120 20 40 60 Time [s] 80 100 120 100 Pin [%] Accelerator 50 pedal stroke 0 [%] In hurry driving situation, the headway distance and velocity errors of simulation results are well-fitted with the experimental data Driver model NAGAI LAB Comparison of average velocity error between the simulation results and experimental data of five drivers The average velocity error t fn eV _ avr = ( V − V ) dt (t fn − t0 ) p f ∫ e V_aver[m/s] t0 Average velocity errors of simulation results are well-fitted with the experimental results Confirms correctness of the identified parameter values and ability of representations of vehicle control process of proposed driver model Driver model NAGAI LAB Comparison of fuel economy between the simulation results and experimental data of five drivers Average error between simulation results and experimental data: 5.1% The proposed driver model can be used to study of driver behavior in a systematical driver-vehicle system Study of Driver Behavior in a DriverVehicle System NAGAI LAB Hv [sm-1] 0.005 0.01 0.015 0.020 0.025 0.055 Hx [m-1] 0.002 0.007 0.012 0.017 0.022 0.055 CV [sm-1] 0.03 Thw [s] t fn eV _ avr = ( V − V ) ∫ p f dt (t fn − t0 ) t0 Bad Good [Hx Hv=0.01]and [HV Hx=0.002] Larger Hx or HV will give better velocity tracking performance The headway distance gain, Hx contributes stronger effect on velocity tracking performance than relative velocity gain, HV Study of Driver Behavior in a DriverNAGAI LAB.System Vehicle NAGAI LAB Hv [sm-1] 0.005 0.01 0.015 0.020 0.025 0.055 Hx [m-1] 0.002 0.007 0.012 0.017 0.022 0.055 CV [sm-1] 0.03 Thw [s] Larger Hx or HV will give better fuel economy Good Bad [Hx Hv=0.01]and [HV Hx=0.002] The headway distance gain, Hx contributes stronger effect on fuel economy than relative velocity gain, HV Drivers who are good sensitive in headway distance tend to give good fuel economy ... Intelligence : to provide service with respect to the operation task  Sensor information is used to recognize the operation modes /behaviour Driver steering behaviour adaptation ・　 lane keeping... Stop sign Brake lamp Vehicle velocity Acceleration Driver state Car following Braking to avoid collision Brake pedal Transition condition Free following Accelerator pedal Vehicle Deceleration... Deceleration at stop line or red signal Decelerating D Stop Assistance Stopping S Stopping Sensor Network of Continuous Sensing Drive Recorder ①Camera/Image Synthesizer Captured VDO image Front Front