This paper describes a genetic algorithm planning method for autonomous robots in unstructured environments. It presents the approach and demonstrates its application to a laboratory planetary exploration problem. The method represents activities of the robot with discrete actions, or action modules. The action modules are assembled into an action plan with a Genetic Algorithm (GA). A successful plan allows the robot to complete the task without violating any physical constraints. Plans are developed that explicitly consider constraints such as power, actuator saturation, wheel slip, and vehicle stability. These are verified using analytical models of the robot and environment. The methodology is described in the context of planetary exploration similar to the NASA Mars Pathfinder mission. More aggressive missions are planned where rovers will explore scientifically important areas that are difficult to reach (e.g., ravines, craters, dry riverbeds, and steep cliffs). The proposed approach is designed for such areas.