Autonomous space rovers face significant challenges when navigating deformable and heterogeneous terrains during space exploration. The variability in terrain types, influenced by different soil properties, often results in severe wheel slip, compromising navigation efficiency and potentially leading to entrapment. This paper proposes SlipNet, an approach for predicting slip in segmented regions of heterogeneous deformable terrain surfaces to enhance navigation algorithms. Unlike previous methods, SlipNet does not depend on prior terrain classification, reducing prediction errors and misclassifications through dynamic terrain segmentation and slip assignment during deployment while maintaining a history of terrain classes. This adaptive reclassification mechanism has improved prediction performance. Extensive simulation results demonstrate that our model (DeepLab v3+ + SlipNet) achieves better slip prediction performance than the TerrainNet, with a lower mean absolute error (MAE) in five terrain sample tests.

We present the reinterpreted and reimplemented Copycat project, an architecture solving letter analogy domain problems. To support a flexible implementation change and rigor testing process, we propose a implementation method in DrRacket by using functional abstraction, naming system, initialization, and structural reference. Finally, benefits and limitations are analyzed for cognitive architectures along the lines of Copycat.