In this project a robot moves in an unknown environment in which it has to reach two basic goals. First of all it has to find a goal marked by a red point. Second the robot has to avoid obstacles of different forms and sizes which are distributed over the whole environment. To achieve this we apply the method of Reinforcement Learning.
Avoiding an obstacle is the basic capability mobile robots have to possess since this enables the robot to move in an unknown or even changing environment without getting damaged. The robot has a differential propulsion which allows to move in an holonomic way. Three different movements are therefore possible: advancing straight ahead, turning right and turning left.
There are basicly two approaches how to take the decision which movement to choose and therewith avoiding obstacles. On one hand it's possible to give the robot some kind of internal map which tells him which decision it has to take when it recognizes a certain point in its environment. This approach needs a huge number of internal states since every possible situation must already been taken. On the other hand there is the reactive approach in which the robot decides only because of his one perception what decision might deliver the best result. Reactive avoiding suit esspecially if the environment is completely unknown or dynamicly changing.
Many algorithms of reactive navigation provide a sufficient simplicity but are rather inefficient since the robot choses a hardly predictable trajectory which not necessarily leads to a given aim. A random movement is the consequence.
This project uses the method of reinforcement learning to navigate in the direction of a goal by efficiently searching a safe trajectory in an environment containing obstacles .
|
|
|