Real-Time Implementation and Validation of Automated Path Following Lateral Control using Hardware-in-the-Loop (HIL) Simulation Poster #1, SAE Technical Paper 2017-01-1683 (doi:10.4271/2017-01-1683) Adit Joshi Ford Motor Company
The current and near-future trend in the automotive industry is towards the increased development of Advanced Driver Assistance Systems (ADAS) and autonomous vehicles to ensure safety of road users and reduce automobile related fatalities. Using vehicle testing only may result in unrealistic and infeasible timescales for software validation of autonomous vehicles. • 100 vehicle fleet • 24/7 518 years of testing 17 billion driven km • 365 days/year • 40 km/hr Trajectory-tracking or path following is a common control problem regarding lateral control for autonomous vehicles. This addresses the need for a control law that can control and actuate a vehicle to follow a reference geometric path. This is critical for autonomous vehicle due to: • 54% of US traffic accidents in 2014 • Lane Departure Warning/Lane (17,791 fatalities) Keep Assist ADAS features – Provide partial lateral control – Ill-timed road departures only – Unintended lane departures – Side-swiping Hardware-in-the-loop (HIL) testing is a testing method which has become an integral part of control validation in the automotive product development cycle due to the following benefits: • Controllers tested in a simulated environment • Scalability and repeatability of scenarios • Improvement in test consistency • Reduction in system variation
0 0
𝑥1 𝑚𝑣 𝑥4 0 0
1 ;;;; 𝑚𝑣 ;;;;0 ;;;;0
Where: 𝑥(𝑡) = 𝛽(𝑡) 𝜓(𝑡) 𝜓(𝑡) 𝑣𝐺 (𝑡) 𝑋(𝑡) 𝑌(𝑡) 𝑦(𝑡) = 𝑋 𝑌 𝑇 𝑢(𝑡) = 𝑆𝑣 (𝑡) 𝐹𝑙𝑅 (𝑡) 𝑇 𝑙𝑓 𝜓 𝑢(𝑡) 𝑆𝑣 = 𝐶𝛼𝑓 (𝛿𝑤 − 𝛽 − ) 𝑣𝐺 𝑙𝑟 𝜓 𝑆ℎ = 𝐶𝛼𝑟 −𝛽 + 𝑣𝐺
-50
-50
-100
𝑇
-250 -243.162 -243.164
𝑢1 = −𝑆ℎ + 𝑚𝑣 𝑥1 𝐶12 − 𝑆12 𝑦1 + 𝐶12 + 𝑥1 𝑆12 𝑦2 𝑢2 = 𝑚𝑣 𝐶12 𝑦1 + 𝑆12 𝑦2 𝑙𝑓 𝑥3 𝑢1 𝛿𝑤 = + 𝑥1 + 𝐶𝛼𝑓 𝑥4
-300
0
RESEARCH POSTER PRESENTATION DESIGN © 2015
www.PosterPresentations.com
-243.165 -250 -243.17
-243.175 972.15
400 600 800 Longitudinal Position of Vehicle (m)
1000
-350
1200
0
Crosswinds (Left) Long. Lat. Traj. Traj. (m) (m) 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
200
972.154
1000
Wind Speed (km/hr)
Crosswinds (Right) Long. Lat. Traj. Traj. (m) (m) 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
Simulation Test Results of Payload Lateral Position for Vehicle Trajectory at 40 km/hr
1:3655 m 1:6386 m 2:1848 m
0 m (Nominal) 0:375 m 0:75 m !0:375 m
-100
-150
-200
-250 -243.1669
-243.167 972.1525 200
-150
!0:75 m
-200 -243.167 -250 -243.1675
-243.168 972.15
972.1528 400 600 800 Longitudinal Position of Vehicle (m)
1000
-350
1200
0
200
Payload Long. Payload Payload Lat. Pos. Lat. Pos. Pos. Long. Lat. Long. Lat. (m) Traj. Traj. Traj. Traj. (m) (m) (m) (m) 0.9999 0.9999 -0.75 0.9999 0.9999 0.9999 0.9999 -0.375 0.9999 0.9999 0.9999 0.9999 0 0.9999 0.9999 0.9999 0.9999 0.375 0.9999 0.9999 0.75 0.9999 0.9999
Payload Long. Pos. (m)
972.154 400 600 800 Longitudinal Position of Vehicle (m)
1.3655 1.6386 2.1848 -
Payload Long. Payload Pos. Lat. Pos. Long. Lat. (m) Traj. Traj. (m) (m) 0.0103 0.0174 -0.75 0.0201 0.0335 -0.375 0.0408 0.0669 0.375 0.75
Simulation Test Results of Payload Mass for Vehicle Trajectory at 40 km/hr
1000
Payload Lat. Pos. Long. Lat. Traj. Traj. (m) (m) 0.0075 0.0097 0.0200 0.0202 0.0340 0.0414 0.0114 0.0136
-50
136 kg (Nominal) -100
544 kg -150
-200 -243.1667 -250 -243.1668
1200
𝑅 =1−
𝑆𝐼 =
2
Where: 2 𝑁𝑖 = Nominal value 1 𝑛 𝑛 𝑉𝑖 = Varied value 𝑖=1 𝑉𝑖 − 𝑛 𝑖=1 𝑉𝑖 𝑛 = Total number of 𝑛 data points 1 𝑁𝑖 − 𝑉𝑖
𝑛
𝑖=1
𝑁𝑖
136 272 408 544
Payload Mass Long. Lat. Traj. Traj. (m) (m) 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
Fric. Coeff. (-) 1 0.75 0.6 0.4
!1 m=s2 (Nominal)
-154.5
!2 m=s2 !4 m=s2 !6 m=s2
7.679 4
6 8 10 12 14 Longitudinal Position of Vehicle (m)
16
18
Accel. Events Long. Lat. Traj. Traj. (m) (m) 0.9973 0.9973 0.9884 0.9884 0.9671 0.9669 0.9672 0.9671
-155.5
360
362 364 366 Longitudinal Position of Vehicle (m)
368
-243.1676
-350
0
200
Payload Mass (kg) 272 408 544
370
Sensitivity Index
Decel. Events Long. Lat. Traj. Traj. (m) (m) 0.9968 0.9968 0.9987 0.9964 0.9393 0.9360 0.9078 0.8984
Accel/ Decel (m/s2) 2 (-2) 4 (-4) 6 (-6)
Accel. Events Decel. Events Long. Lat. Long. Lat. Traj. Traj. Traj. Traj. (m) (m) (m) (m) 0.5033 0.5657 0.0537 0.0396 0.7265 0.8215 0.0585 0.0426 0.7266 0.8240 0.0666 0.0432
CarSim HIL Simulation One-Time Costs Lab $120000 construction Recurring Costs Energy $2000 Installation Labor
$5000 $20000
• Implementation/Validation of automated path following lateral control – Addition of lateral capability to longitudinal controlled HIL setup – Path-following capability of lateral controller validated across different simulated conditions/variations – Repeatability of tests for design of experiments testing in short timeframe – Unsafe driving scenarios tested in safe simulated environment – Sensitivity analyses showcased importance/effects of vehicle/environmental variations for control design
•
-243.1676 -250 -243.1676
400 600 800 Longitudinal Position of Vehicle (m)
1000
1200
Sensitivity Index Fric. Coeff. Long. Lat. Traj. Traj. (m) (m) 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
-154
FUTURE WORK
-243.1676
1000
365.2
-153.5
-200
972.1532 400 600 800 Longitudinal Position of Vehicle (m)
2
-153.5
-155
Recurring Costs Test vehicle $40000 and installation Transportation $20000 Fuel $8000 Staff $180000
-150
-300 -243.1676
Coefficient of Determination
𝑉𝑖 − 𝑁𝑖
0:75 0:6 0:4
-100
408 kg
-243.1669 972.1528
𝑛 𝑖=1
1 (Nominal)
272 kg
-300
2
-4.343
1200
Simulation Test Results of Friction Coefficient for Vehicle Trajectory at 40 km/hr
-50
Payload Mass (kg)
-4.3425
Vehicle Testing One-Time Costs Instrumentation $50000
Sensitivity Index
Coefficient of Determination
200
-4.342 -8
-153
• HIL simulation can address the need for alternative methods for testing automated driving applications – Consistent and controlled test environment – System variables fixed – Application of different disturbance conditions for comparison – Testing and validation costs lower than vehicle testing
-300
0
-6
1 (-1) 2 (-2) 4 (-4) 6 (-6)
Payload Mass and Friction Coefficient Test Results and Sensitivity Analysis:
The simulations of the lateral controller were conducted on the Simulink-CarSim HIL platform representation of 2017 Ford Fusion. The performance of the lateral controller was validated using the measures of Coefficient of Determination (𝑅2 ) and Sensitivity Index (𝑆𝐼) on a virtual track representation of a highway section.
-4
Accel/ Decel (m/s2)
-50
1.0924 1.3655 1.6386 2.1848 -
+6 m=s2
-153.4
CONCLUSIONS
1:0924 m (Nominal)
Payload Long. Pos. (m)
-2
Coefficient of Determination
Crosswinds Crosswinds (Left) (Right) Long. Lat. Long. Lat. Traj. Traj. Traj. Traj. (m) (m) (m) (m) 0.0076 0.0076 0.0137 0.0139 0.0172 0.0234 0.0246 0.0408 0.0259 0.0370 0.0338 0.0550
50 100 150
-50
0
+4 m=s2
-12 0
1200
Sensitivity Index
Simulation Test Results of Payload Longitudinal Position for Vehicle Trajectory at 40 km/hr
-350
+2 m=s
-4.3435 7.678
400 600 800 Longitudinal Position of Vehicle (m)
-153.3 -152.5
2
972.158
Payload Position Test Results and Sensitivity Analysis:
-350
The objectives of this research were as follows: • Implement real-time version of path-following lateral controller – Addition of lateral capability to longitudinal controlled HIL setup • Validate path-following capability of lateral controller • Quantitatively understand real-time behavior and sensitivity of lateral controller using simulations across variations in: – Surface type/friction – Crosswinds – Rapid acceleration/ – Payload position deceleration events – Payload mass
-200
972.158
-300
For controller implementation, the six states ( 𝛽 𝜓 𝜓 𝑣𝐺 𝑋 𝑌 𝑇 ) along with the desired longitudinal and lateral trajectories of the path (𝑋𝑑 , 𝑌𝑑 ) were used as inputs to the lateral controller from CarSim for accuracy reasons.
The simulation of non-hardware subsystem plants and controllers was achieved by using a real-time CarSim-Simulink co-simulation environment representative of the 2017 Ford Fusion Hybrid. A high fidelity plant model of the power-split powertrain comprising an engine, motor-generator, high voltage battery, and planetary gear set driveline was defined in Simulink, which formed the basis of the vehicle level plant model simulation. The Simulink plant model representation also included brakes and steering controllers, along with high voltage battery and auxiliary subsystem models. The brakes, steering, environment, and vehicle dynamics plant models were simulated using the CarSim representation of the 2017 Ford Fusion.
972.154
-100
0 𝑦 + 𝑦𝑑 𝛼12 𝑑
150 km=hr Crosswind (Right)
Coefficient of Determination
0 50 100 150
+1 m=s (Nominal) 0
-10
200
Wind Speed (km/hr)
2
100 km=hr Crosswind (Right) -150
-152
2
-300
-243.168
𝑇
Where: 𝑦1 = 𝜆𝜔1 − 𝛼 𝑥4 𝐶12 − 𝜆𝑥5 𝑦2 = 𝜆𝜔2 − 𝛼 𝑥4 𝑆12 − 𝜆𝑥6 𝑙𝑟 𝑥3 𝑆ℎ = 𝐶𝛼𝑟 −𝑥1 + 𝑥4 1/𝜆 0 𝛼11 𝜔 = 𝑦𝑑 + 0 0 1/𝜆
-100
-243.166
1 −𝑠𝑖𝑛 𝑥1 + 𝑥2 𝑐𝑜𝑠 𝑥1 + 𝑥2 + 𝑥1 𝑠𝑖𝑛 𝑥1 + 𝑥2 𝑠𝑖𝑛 𝑥1 + 𝑥2 − 𝑥1 𝑐𝑜𝑠 𝑥1 + 𝑥2 𝑚𝑣 𝑐𝑜𝑠 𝑥1 + 𝑥2 1 −𝑠𝑖𝑛 𝑥1 + 𝑥2 𝑞 𝑥 = 𝑆ℎ 𝑚𝑣 𝑐𝑜𝑠 𝑥1 + 𝑥2
𝛼 and 𝜆 are tunable parameters in the control law. The control law inputs (𝑢) and the resultant front steering wheel angles (𝛿𝑤 ) are expressed as:
50 km=hr Crosswind (Right)
-243.16
-350
𝑒 = 𝑦𝑑 − 𝑦 = 𝑋𝑑 − 𝑋 𝑌𝑑 − 𝑌 𝑒 + 𝛼𝑒 + 𝜆𝑒 = 0
50 km=hr Crosswind (Lef t)
-200
972.15
From the above law, the control inputs (𝑢) could be determined through the tracking error and tracking error dynamics between the desired trajectory (𝑋𝑑 , 𝑌𝑑 ) and actual trajectory (𝑋, 𝑌) of the center of gravity of the vehicle as shown by:
No W ind (Nominal)
150 km=hr Crosswind (Lef t)
-150
Freund and Mayr also proposed a feedback linearization control law of the form, 𝑦 = 𝑞 𝑥 + 𝑆 𝑥 𝑢, in terms of the vehicle longitudinal and lateral accelerations at the center of gravity where 𝑞 𝑥 and 𝑆 𝑥 were expressed as: 𝑆 𝑥 =
No W ind (Nominal)
100 km=hr Crosswind (Lef t)
Simulation Test Results of Deceleration for Vehicle Trajectory at 40 km/hr
Simulation Test Results of Acceleration for Vehicle Trajectory at 40 km/hr
Lateral Position of Vehicle (m)
0
−
Acceleration/Deceleration Test Results and Sensitivity Analysis:
Simulation Test Results of Crosswinds (Right) for Vehicle Trajectory at 40 km/hr
Lateral Position of Vehicle (m)
1 𝑚𝑣 𝑥4 0 𝑙𝑓 + 𝐼𝑧𝑧
Simulation Test Results of Crosswinds (Left) for Vehicle Trajectory at 40 km/hr
Lateral Position of Vehicle (m)
𝑆ℎ −𝑥3 + 𝑚𝑣 𝑥4 𝑥3 𝑆ℎ 𝑙𝑟 𝑥 (𝑡) = − 𝐼𝑧𝑧 0 𝑥4 𝑐𝑜𝑠 𝑥1 + 𝑥2 𝑥4 𝑠𝑖𝑛 𝑥1 + 𝑥2
SIMULATIONS & TEST RESULTS (CONT.)
Crosswinds Test Results and Sensitivity Analysis:
Lateral Position of Vehicle (m)
INTRODUCTION
The HIL hardware setup of the vehicle level HIL simulation consisted of: • Engine Control Module (ECM) – Conversion of requested powertrain demand into engine speed and torque commands. • Hybrid Control Unit (HCU) – Conversion of requested powertrain demand into generator and motor speed and torque commands. • Gear Shift Module (GSM) – Reception of desired gear position request. • Transmission Range Control Module (TRCM) – Actuation of transmission into desired gear requested from GSM. • Gateway Module (GWM) – Facilitation of communication and transfer of data between various modules.
Freund and Mayr proposed a sixth-order non-linear control-oriented model for path following that adjusts the front wheel steering angles and drive/braking force.
Lateral Position of Vehicle (m)
A HIL simulation provides a test platform where the system under test consists of actual hardware components with the remainder of the system simulated with mathematical or physics-based plant models of the processes via a real-time simulation platform.
SIMULATIONS & TEST RESULTS
Lateral Position of Vehicle (m)
Software for autonomous vehicles is highly complex and requires vast amount of vehicle testing to achieve a certain level of confidence in safety, quality and reliability. According to the RAND Corporation, a 100 vehicle fleet running 24 hours a day 365 days a year at a speed of 40 km/hr, would require 17 billion driven kilometers of testing and take 518 years to fully validate the software with 95% confidence such that its failure rate would be 20% better than the current human driver fatality rate. In order to reduce cost and time to accelerate autonomous software development, Hardware-in-the-Loop (HIL) simulation is used to supplement vehicle testing. For autonomous vehicles, path following controls are an integral part for achieving lateral control. Combining the aforementioned concepts, this research focuses on a real-time implementation of a path-following lateral controller, developed by Freund and Mayr. The controller is implemented on a powertrain subsystem HIL simulation bench to enable lateral control of the longitudinal controlled HIL setup for automated driving applications. 2017 Ford Fusion Hybrid powertrain controllers and actuators were used as the hardware platform for the powertrain subsystem. The simulation of other subsystem plants and controllers was achieved by using a real-time CarSim-Simulink co-simulation environment representative of the 2017 Ford Fusion Hybrid through a dSPACE HIL simulator.
MATERIALS & METHODS
Lateral Position of Vehicle (m)
MATERIALS & METHODS
Lateral Position of Vehicle (m)
ABSTRACT
Payload Mass Long. Lat. Traj. Traj. (m) (m) 0.0146 0.0363 0.0489 0.0785 0.0642 0.1047
Fric. Coeff. (-) 0.75 0.6 0.4
Addition of more controllers and physical systems to HIL such as brakes hydraulics, brakes controller and steering controller. – Ability for further regression testing of controller software/ hardware • Addition of autonomous vehicle target hardware/software – Ability to test actuation and performance of vehicle response
Fric. Coeff. Long. Lat. Traj. Traj. (m) (m) 0.0064 0.0078 0.0065 0.0091 0.0129 0.0150
CONTACT Adit Joshi; Ford Motor Company, Dearborn, MI, USA;
[email protected]