Robotic Localization and Multi-Sensor, Semantic 3D Mapping for Exploration of Subsurface Voids

Abstract:

Robotic exploration of subsurface voids (SSV) has seen rapid advancements in recent years. These advances bring compelling technologies into consideration for the exploration of SSV on other worlds. A critical component of these technologies is 3D Simultaneous Localization And Mapping (SLAM), which provides capabilities for a robot to both know where it is and generate a 3D map of the environment around it. These capabilities are also critical for scientific exploration. Location estimates provide a global context to measurements, identifying how far underground, and how far from the entrance a robot is, to help inform which phase of the cave system it is in. Accurate measurements of a robot’s location are also essential for enabling robotic navigation and obstacle avoidance. SLAM also produces a 3D map of the environment, giving valuable insights into the cave geometry as a product in and of itself, as well as providing context for other measurements with reference to the cave geometry.

There are numerous challenges to achieving accurate 3D SLAM in SSV environments. Orbital imagery cannot be used, there is low or no natural lighting, rough terrain and a lack of persistent fixed landmarks, such as mountains. Our system, Large-scale Autonomous Mapping and Positioning (LAMP) looks to tackle these challenges by leveraging a fusion of multiple sensing modalities for robustness and richness of information. LAMP produces enriched multi-dimensional maps that fuse 3D information with sensing information, such as: surface thermal properties (through IR cameras), ambient measurements (such as gas concentrations), as well as semantic measurements, assessing areas of scientific interest. Lidar is the primary sensor, providing reliable and accurate measurements at large ranges, and is combined with dropped ranging beacons to reduce localization drift. Our full system has been proven for over 1km of traverse with less than 5 m drift in underground environments. In this presentation, we will overview our multi-sensor 3D SLAM algorithm, what it is capable of, and how these algorithms can be extremely valuable future subsurface exploration missions.