Creating a decision-making robot using a neural net

Creating a decision-making robot using a neural net

Zachary J. Tong


Artificially intelligent robots are becoming increasingly important in manufacturing, health care, search and rescue, law enforcement, exploration, and fighting terrorism. Many robots are preprogrammed with a defined set of tasks. Robots that learn from their actions and apply the gained knowledge to solve new problems can be used in more complicated applications. The purpose of this project was to build and program a robot that could learn from its derisions to navigate a path. The hypothesis was to determine if a robot could learn to turn left, right, move forward, and stop based on sensory input and previous decisions stored in a neural net. A Lego[TM]robot was designed with a two-wheel drive and a support wheel. Attached to the front was a sliding rack equipped with a light sensor. A series of colored lines was used as a path for the robot’s light sensor to follow. Two different programs were coded using Not Quite C (an adapted version of C for the Lego[TM]robot). The first program defined a specific set of tasks for the robot to execute. The second program incorporated a neural net. The neural net enabled the robot to test possible solutions to a problem and remember the correct solution. When the robot encountered the next problem, it referenced the neural net. if the correct solution was available, the robot recalled and executed the appropriate action. Otherwise, the robot tested possible solutions to find the correct one. In conclusion, a robot programmed with neural net logic can remember the solutions required to navigate a path. The robot can then apply those solutions to navigate any path regardless of the path’s configuration.


COPYRIGHT 2003 Ohio Academy of Science

COPYRIGHT 2008 Gale, Cengage Learning