Composable Autonomy in Heterogeneous Groups

In multi-robot Simultaneous Localization and Mapping (SLAM), a group of robots explore and map an unknown area. The group can benefit from its size by combining the robots’ maps to improve coverage and by each robot using shared information to improve its own localization. Most approaches to multi-robot SLAM consider homogeneous groups, in which all robots have the same sensors.  But for robots with different sensors, shared observations are difficult to detect. For example, consider a scenario where large ground vehicles with heavy LIDAR sensors map an outdoor region with the aid of small UAVs with only stereo cameras.

Both robots can construct a point cloud from their sensors, but they have very different resolutions and error models, making dense methods for data association. However, correspondences can be found by matching keypoints observed in all types of point clouds.

In this work, we evaluated 5 commonly used 3D keypoint detectors for loop closure in heterogeneous multi-robot SLAM using point clouds constructed from the KITTI dataset. Ideally, a keypoint detector will find repeatable keypoints across different sensor types; find repeatable keypoints in observations made from different poses; detect relatively few high-quality keypoints that can be shared even with low-bandwidth communication; and detect keypoints efficiently from LIDAR and stereo camera point clouds to enable real-time SLAM. Our evaluations showed good performance from two detectors. The NARF keypoint detector is the best choice if computational power or bandwidth is limited, and the KPQ-SI detector found more repeatable keypoints but was much less efficient. To the best of our knowledge, this is the first comparison of 3D keypoint detectors for SLAM or visual odometry

E. Boroson and N. Ayanian, “3D keypoint repeatability for heterogeneous multi-robot SLAM,” to appear in IEEE International Conference on Robotics and Automation, Montreal, 2019.

Task: RA2.A3 Composable Autonomy in Heterogeneous Groups

Points of contact: Nora Ayanian (PI) and Elizabeth Boroson