Autonomous Rock Detection in Planetary Exploration Systems

Autonomous rock detection constitutes a critical capability for planetary exploration rovers and landers, enabling them to identify, classify and navigate around geological features without human intervention. This capability underpins hazard avoidance, path planning and scientific sampling, ensuring mission safety and maximising the scientific return. Key challenges arise from the diversity of rock shapes, textures and scales, as well as variable illumination, dust deposition and sensor noise in extraterrestrial environments. Modern systems integrate multiple sensing modalities—stereo vision, LiDAR, multispectral imaging—and employ a progression of methods from classical edge‐ and region‐based segmentation through machine‐learning classifiers to contemporary deep‐learning approaches. On‐board data processing constraints and real‐time requirements further demand lightweight models optimised for limited compute and power budgets. Recent advances focus on robust feature extraction, adaptive thresholding and context‐aware inference, allowing rovers to distinguish rocks from regolith and shadows. These improvements bolster autonomous navigation, geological mapping and targeted sample collection, contributing directly to astrobiological investigations and resource assessment on Mars, the Moon and beyond.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Autonomous Rock Detection in Planetary Exploration Systems publication trend

The graph below shows the total number of articles in autonomous rock detection in planetary exploration systems across all publications each year (not limited to Nature Index journals).

Technical terms

Convolutional Neural Network: A class of deep learning models that applies convolutional filters to extract hierarchical features from images.

Single-Shot Detector (SSD): An object detection architecture that predicts bounding boxes and class probabilities in a single forward pass.

Semantic segmentation: The process of assigning a class label to each pixel in an image to delineate objects or regions.

Intersection over Union (IoU): A metric measuring the overlap between predicted and ground-truth regions, defined as their area of intersection divided by their area of union.

Transformer: A neural network architecture relying on self-attention mechanisms to model long-range dependencies in data.

Transfer learning: A technique where a model pretrained on one dataset or task is adapted to another, reducing the need for large labelled datasets.

References

1. CNN Based Detectors on Planetary Environments: A Performance Evaluation. Frontiers in Neurorobotics (2020).
2. RockSeg: A Novel Semantic Segmentation Network Based on a Hybrid Framework Combining a Convolutional Neural Network and Transformer for Deep Space Rock Images. Remote Sensing (2023).