Unformatted text preview: ithm.
The algorithm involves finding the shortest distance from source to destination by viewing the entire map in the form of equally divided grids. We begin by marking the start and end positions by capturing an image
of the table with the robot and obstacles. The algorithm first tries to search the shortest path by taking the
travelled distance into account as it uses the Euclidean distance which gives the real cost between the
current and goal states.
In our implementation, we did not consider the paths on the boundaries, corners or edges, as it may make
the robot fall off the table when it executes a turn. The path is computed such that it always takes the
nearest grid to complete the shortest path to the goal state. To make it work more efficiently, valid
heuristics are considered which improves the search performance. The entire algorithm is computed
before the robot actually starts moving as there is no update on the go as it may lower the search
algorithm efficiency when the camera link breaks. Guiding the robot: Given a state we define each possible case and a command to achieve it.
we decided to firstly take the whole scenario, solve the problem and then tell the commands to robot
(maybe because of potential camera failures);
Conclusion:
This work showed us that it is possible to solve a problem that seemed complicated at first sight
with cheap hardware, like a lowcost camera and a lego robot, and simple functions, as
convolution and linear regression. Something that we think that might be changed in future works
is the fact that we do not correct the robot path iteratively, but instead we calculate the path,
send all the commands to the robot and do not update it anymore. The path could be monitored
and corrected from time to time if we had a more reliable camera, it would certainly improve the
results. We believe that the purpose of this project was achieved, since the robot reached the
goal and we were able to learn about the tools we used along the way....
View
Full Document
 Fall '12
 ThomasTrappenberg
 Computer Science, Machine Learning, Sobel, 2 meters

Click to edit the document details