Mindexplore

Addressing the challenge of unreliable long-range planning for robots in exploring unknown planetary surfaces, the dataset simulates unstructured extraterrestrial scenarios using real-world environments, collecting 168 real long-range trajectories (over 38,000 steps, integrating high-precision surface observation data from multi-view RGB and LiDAR), and constructing 37,000 thought chains and command compliance data.

The data covers typical space-based application tasks such as autonomous navigation, dynamic obstacle avoidance, and sample manipulation. By pioneering the introduction of a structured reasoning architecture, the dataset deeply decomposes exploration commands, breaking through the black-box bottleneck of traditional large models and significantly improving the decision-making coherence of long-range tasks. This dataset aims to provide a solid perception-decision-action data foundation for developing highly reliable planetary exploration robots.