Nowadays many public and private institutions begin space studies projects. Among many problems to solve there is a planet exploration. Now rovers are controlled directly from the Earth, e.g. Opportunity. Missions must be planned on the Earth using simulators. Much better will be when the mission planner could set the target area and work to do and the rover will perform it independently. The solution is to make it autonomous. Without need of external path planning the rover can cover a much longer distance. To make autonomous rovers real it is necessary to implement a target leaded obstacle avoidance algorithm. Solutions based on graph algorithms use a lot of computing power. The others use intelligent methods such as neural networks or fuzzy logic but their efficiency in a very complex environment is quite low. This work presents an obstacle avoidance algorithm which uses the genetic path finding algorithm. The actual version is based on the 2D map which is built by the robot and the 2nd degree B-spline is used for the path model. The performance in the most cases is high using only one processor thread. The GA can be also easily multithreaded. Another feature of the algorithm is that, due to the GA random nature, the chosen path can differ each time on the same map. The paper shows the results of the simulation tests. The maps have the various complexity levels. On every map one hundred tests were carried out. The algorithm brought the robot to the target successfully in the majority of runs.