We chose the challenge #1of the BigRed Hacks. We want to improve the sensory system of CUAUV by detecting anomalies in the angle perception. By utilizing cross-category evaluation, we improves the unsupervised anomaly detection and substantially reduces the anomaly cluster. The machine learning module we mainly used is K-Means Clustering. We first use the unsupervised anomaly detection (K Means Clustering) on data like autonomous vehicle's angle over timestamp, velocity over timestamp, depth over timestamp, etc. However, we find that if we only cluster a single variable over timestamp, the anomaly cluster we find is too large and thus meaningless for developer to use to improve the sensory system. So, through plotting two or more variables correlation using Python script, we find a correlation between the vehicle's angle and depth rate: when the autonomous submarine travels on the same plane (depth rate = 0), the angle changes freely; when the submarine travels up and down (depth rate != 0), the angle tends to stay the same at 2rad. We again use the K Means clustering on this correlation and find a much better anomaly cluster. Then, by plotting the angle over timestamp and depth rate over timestamp, we further identify two time periods that are in the anomaly cluster calculated earlier. Because the depth rate is close to 0 during the first time period, we consider the angle anomaly presented here caused by turbulence or other reasons rather than real sensory failure. During the second time period, the depth rate is not closed to 0, so we consider the angle anomaly presented here a possible sensory failure. By using cross-category correlation, we are able to efficiently target the anomalous time period and variable and identify the real sensory failure over anomaly caused by other factors. In the future, we can use One Class SVM and PCA Anomaly Detection module to further prove the anomalies we found in our example. Moreover, adding these two modules in our model all us to distinguish anomalies in future data sets of the sensory system of autonomous vehicles. Lastly, we can apply our method to other robotic machine to analyze anomalies in sensory systems.