One consideration in evolutionary algorithms is how the information being manipulated is represented. For nearly every problem studied in evolutionary computation, there are several methods which could be employed to solve the problem and the choice can greatly affect the results. There may not only be different ways of encoding a problem, but also several representations of the problem within a single coding scheme. This study examines the Tartarus task, a standard grid robot test problem. Different methods of encoding the problem have been studied, but also of interest is the effects of sensor placement on both runtime and overall fitness of the robot. Learning to process sensors has a computational cost that grows combinatorial with the number of sensors, so a decrease in the number of sensors with no decrease in fitness would be desirable. This study shows that how the environment is represented to the Tartarus agent also has a major impact on the quality of the final solution. The results of the baseline, 8-sensor bulldozer can be surpassed by at least two different sensor configurations, both of which have fewer sensors placed strategically.