With the rise of commercial spaceflight and the success of recent asteroid exploration missions, the need for an exploration robot designed to operate in microgravity and rugged terrain has emerged in recent years. Multi-legged autonomous robots with multi-section spike grippers have a decisive advantage over classic, wheeled rovers in terms of mobility in microgravity. In addition, the increasingly lighter and more precise depth imaging sensors on the market have the potential to significantly increase the rover's ability to detect graspable targets in real time.
The study deals with the development and implementation of an algorithm for the detection of graspable targets for the fixation and locomotion of the hexapod rover SCAR-E for terrestrial applications and for planetary exploration. The prototype rover, developed by the Space Robotics Laboratory (SRL) at Tohoku University in collaboration with Asteroid Mining Corporation Ltd. (AMC), utilizes a visual sensor system consisting of six depth cameras to perceive its surroundings. The first development step is to create a high-resolution virtual 3D environment for simulations. By using the Kinect v2 RGB-D camera in conjunction with Real-Time Appearance-Based Mapping (RTAB-Map) technology, a detailed 3D map of a representative rocky terrain is created, which is refined using optimization and post-processing algorithms .
Subsequently, the terrain data obtained is utilized to test the algorithm for detecting graspable targets in the robot's vicinity. Using depth images, the algorithm creates a panoramic point cloud for the terrain surrounding the rover. In the next step, the reference frame of the point cloud is aligned according to the normal of its regression plane, missing parts are added in a Delaunay interpolation and the terrain data is converted into a discrete voxel grid. Based on the geometry of the gripper, a three-dimensional mask is created which represents the gripping extent of the gripper and is used to evaluate each point of the voxelized terrain for its probability to be a graspable target. In parallel, convex areas in the terrain are determined using a curvature analysis and the intersection of the two methods is formed. The algorithm is inspired by the SRL-ClimbLab concept, adapted for SCAR-E and is designed for real-time execution within a ROS/ROS2 framework.
The test results in various scenarios demonstrate the effectiveness of the algorithm in detecting surfaces that can be grasped while taking into account the terrain complexity. The findings confirm the adaptability of the algorithm to the panoramic perception and gripper requirements of SCAR-E and suggest a high reliability and accuracy of the predictions regarding accessibility in rocky terrain. Future improvements aim to optimize speed, enhance the integration of gripper's complete functionality, and improve the detection of steep terrain.
«
With the rise of commercial spaceflight and the success of recent asteroid exploration missions, the need for an exploration robot designed to operate in microgravity and rugged terrain has emerged in recent years. Multi-legged autonomous robots with multi-section spike grippers have a decisive advantage over classic, wheeled rovers in terms of mobility in microgravity. In addition, the increasingly lighter and more precise depth imaging sensors on the market have the potential to significantly...
»