New AI pipeline enables autonomous rock-breaking in mining operations
Real-time RGB-D system runs at 10 Hz on embedded hardware with 675 ms latency.
Impact hammers used for secondary rock reduction in mining are typically teleoperated, limiting efficiency and safety. To address this, Martín Gallegos and colleagues from the Universidad de Chile propose a real-time RGB-D perception pipeline that automatically identifies rocks and generates feasible breaking poses. The pipeline first uses a camera (RGB-D) to capture color and depth data, then applies image-based instance segmentation (likely a convolutional neural network) to detect rocks, and finally processes the point cloud to compute optimal hammer positions while filtering out the robot arm itself.
The system runs on an embedded GPU at approximately 10 Hz with an end-to-end latency of ~675 ms, making it fast enough for closed-loop autonomous control. In experiments using a representative scaled mining environment, the pipeline successfully segmented rocks and generated collision-free poses. This work is a step toward fully autonomous rock-breakers, which could improve mining productivity and worker safety by reducing the need for human teleoperators in hazardous underground environments.
- Pipeline fuses RGB-D instance segmentation with geometric point cloud processing for rock detection and pose generation.
- Achieves ~10 Hz inference on embedded hardware with a total latency of ~675 ms, suitable for real-time control.
- Tested in a representative scaled mining scenario, demonstrating feasibility for autonomous impact hammer operation.
Why It Matters
Autonomous rock-breakers could dramatically improve mining efficiency and safety by minimizing human intervention in dangerous underground operations.