SE 423 Final
Project
Group 74 (2)
Objective
Devise an algorithm to command an autonomous robot vehicle
to travel to five predetermined locations, kill five weeds of either color
(blue and pink), and then report the number of weeds killed in designated
locations. The robot must be able to travel through a course while avoiding
hitting obstacles. Additionally, the robot must be able to recognize the color
of the weeds and report the number of weeds in the zone corresponding to the
weeds¡¯ color. Information about the course such as location of the obstacles
and weeds are to be displayed in LabVIEW.
Path Planning
We
used the A* algorithm to design a path for the robot to traverse the course to
the various way points. As the robot moves through the course, obstacles are
detected and the course map is updated. We created an array of potential
obstacles and when the LADAR sensor detects a point that is close to a possible
obstacle, that obstacle is marked in the map. Depending on the location of the
robot relative to the obstacle wall, we predicted the entire obstacle box and
marked the additional walls. A* is
called again with the updated map and is able to create a path that avoids the
new obstacles. Additionally, we used heuristics that are different from the
homework problem. Instead of Manhattan distance, we used Euclidean distance as
heuristic, which gave us better path in some circumstances.
LabVIEW
LabVIEW
was used to record the path of the robot, the obstacle locations, and the
weeds¡¯ color and corresponding position. The robot¡¯s path is constantly being
updated, and the weed color and position is updated when the robot camera
detects a weed which hasn¡¯t been detected . The wall will
be sent to labview only when it¡¯s first been detected
by LADAR. To make sure the labview picture doesn¡¯t
keep drawing wall and weed, we created an array for weed and wall separately to
keep track whether the wall/weed at current index has been sent or not. We have
developed separate subVIs for obstacle, weed, and
waypoint.
Diagonal Path Implementation
To save the running time of robot in the
course, going diagonally would be a necessary step. In the original A*
algorithm, the next step from current location is only limited to up, down,
left and right. But we have added up-right, up-left, down-right and down-left,
so if there¡¯s no obstacle at up-left corner, and the robot can travel up and
left (no obstacle in up and left direction), then the diagonal nodes will be
pushed in to openset, and the robot could travel in
up-left diagonal direction. Moreover, we¡¯ve modified the Manhattan distance
calculation since the shortest distance is no longer an integer in tiles. If
the next step is in diagonal, then that step distance is 1.414 tiles. Outside
the course, we decide to skip the planned path and let the robot go to the
destination directly, which is pathRow[0], pathCol[0], because there¡¯s
no obstacle outside the course.
Vision Algorithm
The code for Vision Algorithm mostly
reside in two places: First we have set the blue and pink HSV values in the ColorVision.c file so that the camera will constantly
detect blue and pink almost simultaneously. The distance for detecting these
was calibrated and the number of detected pixels to follow a color were also
determined based on the view of the camera. Next, in RobotControl
under the main c file, two robostates were created
where robotstate=1 inspects the the
number of blue/pink pixels, and the blue/pink distances.
Once the robot has detected either a blue
weed or pink weed during its movement to waypoint, it will first check if this weed
has already been detected or not, based on the predicted location of the weed (
based on the current location of the robot and the location of the weed in the
picture). If this is a new weed, and within the course, and the total weed is
less than 5 and the number of each individual color of weed is less than 3,
then the robot will enter robotstate=2. With the
predicted location of the weed, the robot will go there directly and stay for 1
second. After that, it will rerun the A* algorithm to generate a new course map
and enter robotstate 1 because it¡¯s already been
deviated from original path.
Contest Results
First Round : #1
in the contest
Second Round : #2
in the contest
Group Members
Thomas
Liu --- System Engineering and Design 2019
Honglu He --- Electrical Engineering 2019 Yukti Kathuria --- Aerospace Engineering 2018 Xiangyuan (Rocker) Zhang -- Computer Engineering 2020
Source Code: