1. Navi Rutgers University 2012 Design Presentation

2. Mechanical Design Entirely custom chassis Brushed DC drive motors
Designed using SolidWorks
80/20 aluminum framing
0.25” polycarbonate casing
240 lb, including payload
Brushed DC drive motors
80 W, 500 CPR optical encoders
5.6 mph maximum speed
27% maximum grade
Actively air cooled by six fans
100 cfm airflow through chassis
Modeled using CFD simulation

3. Electrical: Power Distribution
Optima YellowTop Battery (×2)
12 V lead acid batteries (in series)
35 A·h capacity
Low power consumption
400 W loaded, 215 W idle
2+ hour battery life
24 V, 12 V, and 5 V DC buses
85%+ efficiency DC-DC regulators
Isolated grounds limit noise
Dashboard
Switches for major components
Dot matrix display status indicator

4. Software Architecture
Use and contribute to open source software when possible
Built on the Robot Operating System (ROS) framework
Three-dimensional Gazebo simulation of driving and sensors
Localization
Perception
Planning
Gazebo
Gazebo

5. Gazebo Simulation

6. Localization: Sensors
GPS: Novatel ProPak V3
2 Hz sample rate
15 cm accuracy (1 sigma)
OmniSTAR HP corrections
Compass: PNI Fieldforce TCM
50 Hz sample rate
0.3° heading accuracy (RMS)
360° tilt correction
Odometry: US Digital Encoders
500 CPR, 0.5 mm resolution

7. Localization: Extended Kalman Filter
Fuse sensors to estimate pose
Odometry: fast, relative pose
Compass: fast absolute orientation
GPS: accurate absolute position
Non-linear “turn-drive-turn” model
(Source: Probabilistic Robotics)
Hardware
Simulation

8. Perception: Sensors Laser: Hokuyo UTM-30LX 40 Hz sample rate
240° field of view
30 m maximum range
Cameras: AVT Manta G-125C (×2)
15 FPS, synchronized
646 × 482 resolution
90° × 65° wide angle lens
130° combined field of view

9. Perception: Laser Field of View
240°

10. Perception: Sensors Laser: Hokuyo UTM-30LX 40 Hz sample rate
240° field of view
30 m maximum range
Cameras: AVT Manta G-125C (×2)
15 FPS, synchronized
646 × 482 resolution
90° × 65° wide angle lens
130° combined field of view

11. Perception: Camera Field of View

12. Perception: Line Detection
Original Image
Color Transformation
Width Filter
Uses the HSV color space to limit the impact of illumination
Width filter is generated from the calibrated camera matrix
Pipelined with left and right images processed in parallel
Total processing time is 100 ms per image pair
Pipelining allows for a 50% increase in sample rate

13. Mapping Local Costmap: (10 m)2 5 cm square cells; high resolution
Always centered on the robot
Used by the local planner
Global Costmap: (1000 m)2
25 cm square cells; low resolution
Origin fixed by a GPS coordinate
Used by the global planner
Sensors mark and clear observations
Based on the ROS navigation stack

14. Planning Global Planner (on demand) Weighted A* Search
Inverse of distance for <1 m
Constant for ≥1 m
Local Planner (20 Hz)
Dynamic Window Approach
10 linear velocity samples
15 angular velocity samples

15. Questions?