a case study of autonomous vehicle: autovalet · 2019-12-19 · a case study of autonomous vehicle:...
TRANSCRIPT
A Case Study of Autonomous Vehicle:
AutoValetDr. Benjamin Ma
Automotive Summit Bangkok
Jun/2019
Agenda
▪ Introduction and Background: the autonomous vehicle program
▪ Auto-parking in modern ADAS
▪ Motivation and Objectives
▪ Approach and Methodology
▪ Commercial application
▪ Results
▪ Q & A
2
Smart Nation: Smart Urban Mobility Initiative
▪ Strategic National
Projects that drive
pervasive adoption of
digital and smart
technologies throughout
Singapore, including
healthcare, living,
mobility and service.
3
Smart Nation: Smart Urban Mobility Initiative
4
▪ Autonomous vehicle technology
is the key enabler to realize
“Smart Urban Mobility”
▪ Smart Urban Mobility is not just
about autonomous vehicle.
▪ Connectivity, shared, assistive
and green
Autonomous Mobility Initiative
5
Gov.
AcademiaIndustry
Autonomous Vehicle in Singapore
6
Singapore Autonomous Vehicle Initiative (SAVI)Joint partnership between LTA, JTC and A*STAR to provide
a technical platform for industry partners and stakeholders
to conduct research and development (R&D) and test-
bedding of AV technology, applications and solutions.
7
Background: a bit of history
▪ Toyota COMS Electric Vehicle
▪ Modified for drive-by-wire/steer-by-wire
8
Background: a bit of history
System Integration: 6 months
3D SLAM on a robot
9
Background: Progress
▪ 3D mapping and localization
▪ Planning
▪ Perception
▪ Longitudinal (velocity) and lateral (steering) control
▪ Sensor fusion
▪ Campus navigation
10
What’s Next?
▪ No COE, no R&D license plate
▪ Not road ready
▪ Autonomous vehicle test site: Fusionopolis and NTU CleanTech park
▪ Limited resources and change in focus area
11
What’s Next?
▪ Drivers in London typically spend almost eight minutes searching for somewhere to park after each journey.
▪ People in KL Waste 25 Minutes Daily Just to Look for Parking Spots.
▪ People even spend more time walking from carpark to their places.
▪ Auto-parking: Driverless Valet Parking, Automated Valet Parking
Telegraph.co.uk
12
Auto-Parking in Modern Commercial ADAS
▪ First commercial auto-parking: Toyota Prius, 2003, automatic parallel parking function (Intelligent Parking Assist)
▪ 2006, Lexus LS, parallel and angle parking
▪ After 2009, Ford, BMW, Audi, etc.
▪ One of the option of modern ADAS
▪ Passive (parking assistance) and Active types (auto-parking)
The British Toyota Prius with Intelligent
Parking Assist has a dashboard screen
to tell the driver what to do.
13
Auto-Parking in Modern Commercial ADAS
▪ First commercial auto-parking: Toyota Prius, 2003, automatic parallel parking function (Intelligent Parking Assist)
▪ 2006, Lexus LS, parallel and angle parking
▪ After 2009, Ford, BMW, Audi, etc.
▪ One of the option of modern ADAS
▪ Passive (parking assistance) and Activetypes (auto-parking)
▪ Human-in-the-loop
Front
Sensor
Rear
Sensor
Side Sensor
14
▪ To develop an auto-parking system including hardware, software and algorithms.
▪ The auto-parking system must be integrated into existing auto-driving framework
▪ The car must be able to locate, navigate to and park itself into the carpark without human intervention. (human-not-in the-loop)
▪ Avoid any static/dynamic obstacles during parking
▪ Scenario: driver simply choose a carpark lot and the car shall be able to park by itself (end-to-end).
Motivation and Objective:
15
Approach and Methodology
▪ Vehicle Modelling
▪ Path planning
▪ Integration of auto-parking into existing autonomous driving framework
▪ Ackermann steering geometry
16
Vehicle Modelling:
l
o i
𝑟
Centre of
turning circlei
m
𝑙
i
Front-back wheel centre distance
(wheelbase)
i Actual vehicle steering angle
Turing radius
m
sa
𝑟
𝑙
Left (outer) wheel orientation angle
Right (inner) wheel orientation angle
Left-right wheel centre distance (track)
▪ Bicycle Model
Centre of
turning circle
𝑟
𝑙
i
i
= tan-1( l/r)o
Steering angle
Wheelbase
Turning radius
17
Width 1.40m
Length 2.406m
Height 1.505m
Steer angle [-660,660]
degree
Steering ratio 21.3
Front wheel angle [-31,31]
degree
Physical property of our NYP car
Vehicle Modelling:
v:
Angular velocity
Linear velocity
w:
(Kinematic
constraint 1)
r >=2.55minimum turning radius
|v/w|>=2.55
(Kinematic
constraint 2)
Equation of
steering angle
conversion
Used in Path Planning part Used in Control part
sa
𝑙
Vehicle model structure
= tan-1(l*w/v)sa
Derive
r =v/w
Angular velocity
Linear velocity
Turning radius Steering angle
Wheelbase
Turning radius
sa = tan-1( l/r)
Equation1: Equation2:
18Timed-Elastic-Bands
Path planning development:
1. Subject to kinematic constraints
2. Planning for car-like robot/mobile base
3. Ability of avoiding dynamic obstacles
4. Human-like behavior for auto-parking path
5. Fast
Requirements of path-planner for car:
1. Search-based: A*, RRT, probabilistic roadmap
2. Reward-based: potential field
3. Sampling-based: SBLP
Motion/Path planning methods
Elastic Band Planning Methods
1. Use global planner to plan a global path (A*,
D*, etc.)
2. Apply two forces: an internal contraction and an
external repulsive force to modify the global
path like an “elastic band”
3. Dynamic obstacles, kinematic constraints,
environmental constraints can be
configured/encoded in the “force”.
19Timed-Elastic-Bands
Elastic Band path planner
Elastic Band Avoid an obstacle
This path should satisfy our car kinematic constraints:
• R>=2.55m• v/w>2.55m• Wheelbase=1.53m
20
Simulator
Setup of simulation environment:
Visualized carpark
Car FootprintLaser scan
Car
Obstacles
Other parked cars
Visualization of path-planning and control
Planned path
21
Desired planed path Undesired planned path
Problem revealed in simulation
Conclusion: Elastic Band planner can’t always plan a desired path for auto-parking
(Human-like)
22
First position: plan a path to navigate the car close
enough to the parking lot
Second position: plan a path for the car to park
inside the parking lot
Addressing the problem: Divide-Conquer Strategy(Just like human)
Computing the first position: trajectory rollout
Roll out a trajectory (kinematic equations) from the second
position to the first position for checking collisions
Increase the turning radius of projected trajectories
until there is no collision and minimum clearance
Sample a random points from the candidate points
No
collision
Obstacle
24
Approach and Methodology
▪ Vehicle Modelling
▪ Path planning
▪ Integration of auto-parking into existing autonomous driving framework
▪ Costmap
▪ ROS concept for space quantization and motion planning
▪ Global and local
▪ 2D/3D
▪ Configurable: inflation, update rate, etc.
▪ Carpark lot representation in map
▪ Assumption: carpark location is fixed in the map
▪ Size: proportional to the Toyota COMS electric vehicle
27
Elastic
Band
Planner
Linear
Velocity input
Angular
Velocity input
Throttle commandBrake command
Steer command
PID velocity
control
Steering angle
control
Vehicle controller
(Actual velocity feedback)
(Actual steering angle feedback)
= tan-1( l*w/v)saFrom Vehicle Modelling part:
How was it integrated into a driverless car?
Longitudinal
Lateral
Localization
Perception
costmap
vehicle
6 DOF state
General
Navigation
Planner
28
Commercial application: Valet auto-parking
Integrate autonomous navigation and auto-parking into one framework
29
Thank you!!
30
31
Some outreach and public demos
ITS Summit 2017 LEAP Exhibition 2017