System Integration and Intelligence Improvements
for WPI’s UGV - Prometheus
Craig DeMello (RBE), Eric Fitting (RBE/CS), Sam King (RBE),
Greg McConnell (RBE), Mike Rodriguez (RBE)
Advisors: Professor Taskin Padir (ECE/RBE) and Professor William Michalson (ECE/RBE)
Abstract
Control Architecture
This project focuses on realizing a series of operational
improvements for WPI’s unmanned ground vehicle
Prometheus with the end goal of a winning entry to the
Intelligent Ground Vehicle Challenge. Improvements include a
practical implementation of stereo vision on an NVIDIA GPU, a
more reliable implementation of line detection, a better
approach to mapping and path planning, and a modified
system architecture realized by an easier to work with GPIO.
The end result of these improvements is Prometheus has
improved autonomy, robustness, reliability, and usability.
Background
Prometheus is an ongoing project at WPI, with the project goal
centered around developing a robot capable of being competitive at
the international Intelligent Ground Vehicle Competition.
2010
Figure 10: Robot path planning
Figure 4: 2011 Architecture
Figure 5: 2012 Architecture
Goals:
• Exchange cRIO for Arduino
• Move peripherals to computer
Benefits:
• More robust system
• Drastically reduced programming time
2012
2011
Path Planning
Using its sensory information,
Prometheus can intelligently plan paths.
• The red outline is the robot footprint
• The blue line is the path from the current
position to the goal
• The red line is the short term path based
on the blue line and tentacles
Localization
Prometheus’s Extended Kalman Filter:
• Integrates GPS, Wheel Encoders and
Compass
• Reduces error from sensor drift
• Provides an accurate estimate of absolute
position and heading
Figure 11: EKF Results
Obstacle Detection
Stereo Vision
Promethus can detect obstacles using it’s two cameras and
stereo vision
Figure 1: 2010 robot
 Mechanical
Construction
 Tentacles
Implemented
 Did not qualify
 Rookie of the
year award
 ROS
Implemented
 Dual cameras
and DGPS boom
mounted
 DGPS and Lidar
moved to main
computer
 Qualified, 13th
place in
navigation
challenge
Figure 12: Right stereo camera
Figure 3: 2012 robot
Figure 2: 2011 robot
 Use 5% of pre
existing code
 Stereo vision
 EKF
 Wheel Encoders
added
 Motors and
Compass to main
computer
 Peripherals to
Arduino
Figure 6: Robot point of view
Line Detection
Goals
Develop a robot capable of being competitive in the
national IGVC competition with the ability to:
Autonomously navigate to GPS waypoints
Avoid all obstacles
Avoid crossing any painted white lines
Navigate around flags based on their color
Figure 7: Virtual map
The world as Prometheus sees it using its SICK Lidar
• The green outline represents the robot footprint
• The red grid cells represent obstacles
• The yellow cells represent virtual inflation, to keep the robot a safe
distance from obstacles.
Figure 8: Robot view of line
•
•
•
•
The shade
of grey
determines
the
distance to
the item in
view
Figure 9: Line in virtual map
Prometheus has the capability to determine
the existence and location of lines it cannot
cross by:
1. Morphing the image from the camera into a
birds eye view
2. Blurring the image
3. Filter out the lines using a Hue Saturation
Value filter and pre determined values
4. Running a Hue lines algorithm
Figure 13: Left stereo camera
Figure 14: Stereo vision disparity map
Accomplishments
 Robust modular system
 Competitive performance
 Implementation of new features over last year
Recommendations:

Smaller, differential drive platform
Sponsors:
References
2010 MQP Report: Design and Realization of an Intelligent Ground Vehicle
2011 MQP Report: Realization of Performance Advancements for WPI’s UGV – Prometheus