Vehicle Lateral Control Under Fault in FrontRear Sensors

1
Vehicle Lateral Control Under Fault in Front/Rear
Sensors
Jihua Huang, Graduate Student Guang Lu, Graduate
Student Masayoshi Tomizuka, Professor Department
of Mechanical Engineering University of
California Berkeley, CA 94720
T.O. 4204
2
Schematic of Lateral Control System in PATH
Vehicles
3
Motivation and Objectives

• Motivation Develop and implement degraded mode
  lateral control schemes to realize fail-safe
  operation
• Specific Objectives
• Investigation of lateral control with information
  from only the front or the rear set of
  magnetometers
• Investigation of transition behavior between
  normal and degraded mode control strategies
• Investigation of autonomous vehicle following
  control in the lateral direction

4
Vehicle Lateral DynamicsInputs ? (wheel
steering angle) ? (road curvature)Output ys
(sensor output)

• V(s) varies with
• vx longitudinal velocity of vehicle
• ds distance between vehicles cg (center of
  gravity) and lateral error sensor

5
Vehicle Lateral DynamicsFront Wheel Steered
Single Unit Vehicle
effects of vx (20, 30, 40m/s)
effects of ds (15m)
300
200
100
0
-100
40
ds
20
0
To Y(1)
-20
-40
-60
10

• -4

10

• -3

10
-2
10

• -1

10

• 0

10

• 1

10
2
6
Principle of Geometric Look-Ahead
7
Front Magnetometer Based Control

• Front Magnetometer Based Control
• Controller has been tested up to 84 mph without
  preview at Crows Landing (test vehicle Buick
  LeSabres)
• Good lane-keeping performance, smooth steering
  action, minimal oscillations
• Transition from nominal control to the front
  magnetometer-based (FMB) control
• The transition behavior mainly depends on the
  performance of the controllers.
• Without delay, good performance under each of the
  nominal and FMB controllers guarantees
  satisfactory transition behavior.

8
Rear Magnetometer Based (RMB) Control

• Characteristics of V(s) (with rear magnetometers)

• V(s) varies with vx (the longitudinal
  velocity).

9
RMB Controller Design (I)Feedback linearization
with unmatched observer

• Basic ideas
• Feedback linearization is a powerful tool to
  design feedback controller for nonlinear or time
  varying systems
• However, the non-minimum phase nature of the
  vehicle lateral dynamics makes a routine
  application of feedback linearization
  inappropriate (state feedback control or matched
  observer state feedback control causes unstable
  internal dynamics)
• An unmatched observer, instead of a Luenberger
  observer, has been designed to overcome the
  difficulty

10
RMB Controller Design (I)Feedback Linearization
with Unmatched Observer

• Feedback linearization
• Time varying (nonlinear) dynamics
• Desired closed loop dynamics
• Control input

Given by a mismatched observer.
11
RMB Controller Design (I)Feedback Linearization
with Unmatched Observer

• Design objectives of the unmatched observer
• To make the closed loop dynamics under unmatched
  observer free of unstable internal dynamics i.e.
  to achieve stability of the closed-loop control
  system.
• To obtain accurate state estimates of the vehicle
  lateral dynamics, such that the closed-loop
  dynamics will be close to the desired closed-loop
  dynamics under feedback linearization
• H-infinity based technique has been used in the
  observer design

12
RMB Controller Design (II) Model-matching Method

• Basic Idea Use the FMB control system as a
  reference model in the H-infinity based RMB
  controller design

13
RMB Controller Design (II)Model-matching Method

• Simulation Results (v 10, 20, 30, 40m/s)

14
Autonomous Vehicle Following Control(in Lateral
Direction)

• Controlled vehicle follows the trajectory of a
  lead vehicle
• Steering command is computed based on the
  information about the relative position between
  two vehicles

A Video Taken at Richmond Field Station
15
Characteristics of Vehicle Following(in
Comparison with Road Following)

• Autonomous Vehicle Following (AVF) does NOT rely
  on any road infrastructure, e.g. the magnetic
  reference system
• AVF relies on measurements of the relative
  distance between the lead and following vehicles
• Lead vehicle is under manual driving or road
  following control
• Sensing is an important issue
• In AVF, the lead vehicle strongly affects the
  following vehicle
• AVF and magnet-magnetometer-based lateral control
  may be combined for enhanced safety and
  reliability.

16
Motivations and Objectives(For Vehicle Following
Control)

• Develop a back-up system in case of complete
  failure of magnet-magnetometer-based automated
  steering control
• Develop an assisting system for degraded-mode
  steering control to accommodate partial
  magnetometer failure (failure in only one set of
  magnetometers)

17
Laser Scanning Radar Sensor (LIDAR)
18
Laser Scanning Radar Sensor
Object
Pulse of Laser beam
Reflection
Laser beam
Reflection
Horizontal scanning
Flight time
Laser Diode
Receiver
Distance Flight Time ? Speed of Light
19
Lead Vehicle
20
Measurements of Relative Position

• Direct measurements from LIDAR (for every 0.1
  sec) 80 sets of
• Distance (resolution 0.15m)
• Angle (resolution 0.15o)
• Intensity (integers from 0 to 31)
• Question How do we distinguish the target from
  surrounding clutter working on the measurements
  from LIDAR?

21
Probabilistic Data Association
measurements
x
x
x
x
x
LIDAR
?i
predicted measurement
x
X
x
validation gate
x
x
x
22
Probabilistic Data Association (Summary)

• Describe the target motion by a kinematic model
• Estimate and predict the target position by a
  Kalman Filter
• Consider all measurements within a validation
  region about the predicted point and associate
  each measurement a probability of being correct
  measurement (intensity is included in the
  probability function)
• Update the Kalman Filter by the weighted sum of
  vi

23
Vehicle Lateral Dynamics (For Vehicle Following
Case)
d disturbance caused by dynamics of lead
vehicle
Bode Plot of V(x), L 816m, Vx 8m/s
24
Tracking Methods

• Minimizing lateral error yL (LIDAR)
• Trajectory-based tracking (LIDAR, yaw rate
  sensor, velocity sensor)
• Tracking with communication (LIDAR, communication
  equipment)
• Tracking with combined use of LIDAR and
  magnetometers

Bode plot of controller



25
Experiment Following an Automatically-Steered
Vehicle

• Experimental setup
• Lead vehicle
• longitudinal manual control (driver controls
  velocity)
• Lateral automatic control (magnetometer-based
  control)
• Following vehicle
• longitudinal manual control (driver controls
  velocity)
• Lateral automatic control (LIDAR-based
  autonomous vehicle following)

Maximum Speed 20 MPH Longitudinal Spacing
about 10 m (controlled manually)
26
Experimental Results
Following Vehicle
27
Experimental Results
Following Vehicle
28
Conclusions

• Rear Magnetometer-Based Control
• Feedback linearization with an unmatched observer
  eliminates the time varying terms approximately
  and overcomes the difficulties posed by the
  non-minimum phase nature of the rear
  magnetometer-based lateral dynamics.
• Model-matching method utilizes the FMB control
  system as a reference model, thereby, avoiding
  the difficulty in choosing appropriate weighting
  functions in direct H-infinity based control.
• Simulations indicate that the controllers can
  provide reasonable performance. Experiments are
  currently undergoing.

29
Conclusions

• Autonomous Vehicle Following
• LIDAR sensor has been incorporated in the vehicle
  control system.
• Autonomous vehicle lateral following controller
  has been tested at low speeds.
• Performance has been evaluated based on
  magnetometer measurements.
• Lane-keeping operation has been achieved at
  speeds up to 20 MPH
• The following vehicle has larger tracking errors
  than the lead vehicle.

30
Work in Process

• Experimental validation for
• Rear magnetometer-based control algorithms
• Autonomous vehicle lateral following controllers
• Trajectory-based tracking
• Tracking with communication
• Tracking with combined use of LIDAR and
  magnetometers