Learning Capabilities Imposed on Low Cost Autonomous Robot

The project uses a DVD box skeleton in order to build an autonomous robot based on PIC18F2620 microcontroller containing 1K EEPROM and 64K flash memory.

Several sensors are used by the robot to read environmental data including IR LED based proximity sensors and light sensors. Through the TC4427A MOSFET drivers’ control, motion is produced by using two DC motors. It can be based on reflex behavior as it moves towards light or can learn to move by using Q-learning algorithm. Each time an increase in light intensity is received by the robot, a reward is being given by the algorithm. When the robot chooses to move straight instead of spinning, it also gets a reward. This illustrates the learning capabilities in a simple neural network.

The sensors are mounted on independent boards, making them removable for testing, maintenance, and other usages. The photoresistors are used to build the 4 adjustable sensors. The whole system is powered by 4 AA batteries and if motors are used, more power might be allocated. The programming, debugging, reading and writing the content of EEPROM are enabled by connecting the motherboard to In-Circuit Serial Debugger.

Read more