Neuromorphic Vision-Based Dataset for Soil Characterization and Slip Estimation in Rover Applications

Advancements in the field of robotics boosts planetary and extraterrestrial explorations, yet rover mobility remains a challenge in diverse terrain conditions with high chances of the rover getting stuck or suffering wheel slip causing deviations from the planned path of the rover. Most rovers rely on frame-based cameras which exhibit considerable limitations like high power consumption, extensive storage demands and degraded visual clarity under low light conditions. To address these limitations, we introduce a novel rover perception approach using an neuromorphic camera which offers microsecond latency, high temporal resolution, and low power consumption. A neuromorphic vision-based dataset is presented that will significantly automate two fundamental navigation tasks such as soil characterization and slip estimation. The dataset comprises 270 experiments collected over five different speeds, two lighting conditions, three soil types and two failure scenarios, capturing wheel-terrain interactions. The dataset is a benchmark for event-based perception and facilitates reproducibility through custom models that can be trained and validated on the open-source data for real-time soil characterization and slip estimation.