Oregon WAVE DARPA Grand Challenge Team

1
Oregon WAVE DARPA Grand Challenge Team

• Matthew MacClary
  (macclary_at_engr.orst.edu)
• Master's student in ECE

2
Oregon WAVE DARPA Grand Challenge Team

• What is the Grand Challenge?
• Our approach to the challenge.
• Software / firmware design.
• Fuzzy logic steering controller.
• Fuzzy terrain planner.
• How to get involved.

3
Grand Challenge 2005

• 2nd (and maybe the last) ground robot race with
  no human control.
• DARPA organized (Defense Advanced Research
  Projects Agency).
• 175 mi. max, with no human control.
• Secret desert route, no human intervention
  allowed.
• October 8, 2005 robots drive the course with no
  human direction at all.

4
Oregon WAVE Approach

• Vehicle SAE mini-baja champion off-road race car
  built here at OSU.
• Team anyone from the Willamette Valley or
  anywhere else really.
• Sponsorship OSU, Mechanical Eng., OSU's SAE,
  Mentor Graphics, HP etc
• Sensing 2x SICK LADAR.
• Computing 4x 16MHz AVR microcontrollers, 1x
  75MHz Xilinx MicroBlaze soft processor. No
  moving parts, programmable in C.

5
(No Transcript)
6
Firmware Design
GPS Board
Steering Board

• GPS coordinates
• GPS/INS sensor
• communication
• Throttle controller

• LADAR input
• Terrain planner
• Steering controller

Servo Board
Engine Board

• Steering position

• Brake actuation
• Throttle actuation
• Engine starting

H-Bridge Board

• Brake light
• Siren
• Fuel pump

7
Fuzzy Intelligence

• Fuzzy logic lets the programmer and the computer
  speak the same language.
• Fuzzy logic statements can be evaluated very
  efficiently by a computer, comparable to
  evaluating Boolean logic statements.
• Only integer arithmetic is necessary, floating
  point math is not required.

8
Fuzzy Steering Servo

• Servo means to presicely position something
  (such as a steering wheel) using active feedback.
• If the wheels are pointed the wrong way, turn
  them toward the right way.
• If the wheels get close to the correct position
  and are going to overshoot, then stop moving the
  wheels!

9
Linguistic Variables

• For the computer to understand what a steering
  servo is, we have to define the terms we are
  using
• Wrong way - wrong_way
• Close to the correct position - close
• Going to overshoot - too_fast
• IF wrong_way AND NOT (close AND too_fast)
• As code

steering and(wrong_way, not( and(close,
too_fast) ) )
10
Linguistic Variables

• When is the term wrong_way true if you are
  driving a car?

wrong_way trueness
100
100
11
Linguistic Variables

• When is the term wrong_way true if you are
  driving a car?

wrong_way trueness
100
50 ???
100
12
Linguistic Variables

• When is the term wrong_way true if you are
  driving a car?

wrong_way trueness
100
25 ???
100
13
Linguistic Variables

• When is the term wrong_way true if you are
  driving a car?

wrong_way trueness
100
10 ???
100
14
Linguistic Variables

• When is the term wrong_way true if you are
  driving a car?

wrong_way trueness
100
1 ???
100
15
Linguistic Variables

• How about an efficient linear approximation?

wrong_way trueness
100
100
16
Linguistic Variables

• Let's define close in a similar fashion

close trueness
100
100
17
Linguistic Variables

• The too_fast variable looks at the speed with
  which the wheels are moving.

too_fast trueness
100
100
18
Logical Operations

• Now that we have defined our linguistic variables
  we need operators that can use them.

Fuzzy AND Truth Table
19
Logical Operations
Fuzzy AND Truth Table
20
Logical Operations
Fuzzy AND Truth Table
21
Logical Operations
Fuzzy AND Truth Table
22
Logical Operations
Fuzzy AND Truth Table
23
Logical Operations
Fuzzy AND Truth Table
24
Logical Operations
Fuzzy AND Truth Table
25
Logical Operations
Fuzzy AND Truth Table

• Fuzzy AND returns the minimum.
• Fuzzy OR returns the ________?

26
Logical Operations
Fuzzy AND, OR Truth Table

• Fuzzy AND returns the minimum.
• Fuzzy OR returns the maximum.
• Fuzzy NOT returns the ________?

27
Logical Operations
Fuzzy NOT Truth Table

• Fuzzy AND returns the minimum.
• Fuzzy OR returns the maximum.
• Fuzzy NOT returns the difference from 100.

28
Fuzzy Controller

• Apply the fuzzy operators

steering and(wrong_way, not( and(close,
too_fast) ) )
29
Fuzzy Controller

• Apply the fuzzy operators

steering and(wrong_way, not( and(close,
too_fast) ) )
30
Fuzzy Controller

• Recall that close starts becoming true when the
  error gets to about 80

close trueness
100
100
31
Fuzzy Terrain Planner

• A desirable route for any vehicle will need to be
  free from obstacles and travel toward the
  destination.
• Lets look at the maximum width of vehicle that
  could pass by an obstacle without making contact

Obstacle 1
Obstacle 2
Vehicle
32
Fuzzy Terrain Planner

• A desirable route for any vehicle will need to be
  free from obstacles and travel toward the
  destination.
• Lets look at the maximum width of a vehicle that
  could pass by an obstacle without making contact

Obstacle 2
A 2 meter wide vehicle could just make it.
Vehicle
33
Fuzzy Terrain Planner

• Our vehicle is about 1.5m wide, so we can choose
  2.5m as the minimum space we want to drive
  through

wide trueness
100
6m
20m
2.5m
34
Fuzzy Terrain Planner

• We want the vehicle to travel directly to the
  next waypoint, but we don't want the vehicle to
  stop driving if it finds a dead end!

direct trueness
100
200
35
Fuzzy Terrain Planner

• Quality at a specific heading is the fuzzy AND of
  wide and direct

36
Fuzzy Terrain Planner

• A basic terrain planning problem is driving
  through a gate. DARPA included a gate at the NQE

37
Fuzzy Terrain Planner

• The fuzzy terrain planner sees the opening in the
  gate as the best route.

38
Fuzzy Terrain Planner

• Successive frames of LADAR data are filtered with
  an averaging filter so that the terrain planner
  is very resistant to noisy data.
• The terrain planner can still find the best path
  with only 25 good data sets.

Running average of route quality determines
vehicle's heading.
39
Get Involved!

• Mailing lists at lists.oregonstate.edu, search
  for
• grandchallenge
• grandchallenge_dev