evaluate path planner and vehicle controller for automated parking · 37 evaluate path planner and...

1© 2019 The MathWorks, Inc.

Evaluate Path Planner and Controller for Automated

Parking

Shusen Zhang

Application Engineering, MathWorks

2

Learn about path planning with these examples

MATLAB

example

Simulink test

bench

3

Learn about path planning with these examples

How robust is the

algorithm?

How can I handle

moving pedestrian

How can I automate

the tests?

4

Evaluate Path Planner and Controller for Automated Parking

▪ Explore system robustness with simulation

▪ Improve design to handle moving pedestrian

▪ Test automation for regression tests

5

Requirement 1: vehicle can only move forward

6

Requirement 2: vehicle can’t cross parking lanes

7

Requirement 3: algorithm must be able to handle loop.

8

Baseline model in the product

9

Single test case in the baseline model

10

Software model

11

Main modules

Planning

Inflation

Collision

Checking

12

Main modules

PlanningControlPath

Analyzer

13

Explore baseline behavior with multiple goal poses

14

Explore behavior with single goal pose

15

Explore behavior with single goal pose

16

Changing testing condition

17

Explore behavior with goal pose that generates a loop in path

18

Isolate issue based on simulation results

Planner

finds the

goal

Controller

is mostly

tracking

Path

Analyzer

19

Existing path analyzer

▪ Find a point on the path for the vehicle to follow

Minimum

distance to entire

path

20

Existing path analyzer

21

Modified path analyzer

▪ Find a point on the path for the vehicle to follow

Incremental

projection

Ref Path

i

i+1

i+2

i-1

u

u = 1

22

Modified path analyzer

23

Explore behavior with improved path analyzer

24

Reduce turning radius and speed

25

Evaluate Path Planner and Controller for Automated Parking

▪ Explore system robustness with simulation

– Remove intermedia points

– Specify different parking maps and spots

– Identify design issue and improve the design.

▪ Improve design to handle moving pedestrian

▪ Test automation for regression tests

26

Build test bench to test dynamic scenario

Scenario Authoring

27

Identify changes to react to pedestrian 1. Update map

based on

scenario

2. Aware of

potential

collision3. React to

imminent

collision

28

Update map based on scenario1. Update map

based on

scenario

2. Aware of

potential

collision3. React to

imminent

collision

29

Update map based on scenario

30

Identify changes to react to pedestrian 1. Update map

based on

scenario

2. Awareness of

potential

collision3. Action with

potential

collision

31

Awareness of potential collision

Time-to-

collision

calculation

32

Time-to-collision with known path

Reference

Actual

33

Identify changes to react to pedestrian 1. Update map

based on

scenario

2. Awareness of

potential

collision3. Action with

potential

collision

34

Action with potential collision

▪ Reduce speed when TTC is low

Time to collision

Input

35

Test in dynamic environment

36

Improve design to handle pedestrians

37

Evaluate Path Planner and Controller for Automated Parking

▪ Explore system robustness with simulation

– Remove intermedia points

– Specify different parking maps and spots

– Identify design flaws and improve the design.

▪ Improve design to handle moving pedestrian

– Add moving pedestrian

– Create costmap from ground truth

– Reduce speed based on time-to-collision

▪ Test automation for regression tests

38

Automate regression testing

39

Test assessment metrics

40

Test assessment metrics

Trajectory

Reference

PointDistance

Metric

Yaw Metric

TTC Metric

Pedestrian

41

Test iteration

▪ Tests definition and test management

42

Testing multiple goal poses

Pedestrian

43

Automate regression testing

▪ Use test manager to inspect reason of failed test. Yaw Metric

44

Automate regression testing

Less deviation

with higher

gain

45

Automate regression testing

Re-planning

and latency

Environment

model

uncertainty

Complex

parking lot

Backward

motion

46

Automate regression testing

……………………….

47

Evaluate Path Planner and Controller for Automated Parking

▪ Explore system robustness with simulation

– Remove intermediate points

– Specify different parking maps and spots

– Identify design flaws and improve the design

▪ Improve design to handle moving pedestrians

– Add moving pedestrian

– Create costmap from ground truth

– Reduce speed based on time-to-collision

▪ Test automation for regression tests

– Add metrics for planner and controller

– Add test case definition/ management