{"id":353,"date":"2020-04-03T04:03:53","date_gmt":"2020-04-03T04:03:53","guid":{"rendered":"http:\/\/mrsdprojects.ri.cmu.edu\/2020teami\/?page_id=353"},"modified":"2020-05-08T20:48:09","modified_gmt":"2020-05-08T20:48:09","slug":"slam-subsystem","status":"publish","type":"page","link":"https:\/\/mrsdprojects.ri.cmu.edu\/2020teami\/slam-subsystem\/","title":{"rendered":"SLAM Subsystem"},"content":{"rendered":"

Spring Semester Implementation Details<\/h3>\n

Simulation<\/h4>\n

Milestone For SVD<\/h5>\n
\n
\n
We are currently using the real-time appearance-based mapping (RTAB) algorithm for semi-dense mapping. RTAB was chosen due to its versatility in sensor combination and its ability to take in odometry as input. Initially, the robot\u2019s current state was estimated from fusing the wheel odometry with the filtered IMU data using an extended Kalman Filter. This approach was later dropped due to erroneous wheel odometry generated due to wheel slip in both simulation and hardware. Currently, odometry is generated using the ICP algorithm which registers current pointcloud with the original map cloud. It also utilizes the IMU data as an initial estimate. The point clouds are preprocessed for better registration. The point cloud from the depth camera and the 3D lidar are fused together by a custom node to better register objects closer to the robot. We also perform RGB to depth registration using the rgbd_sync nodelet of RTAB. A few parameters in the configuration of RTAB that were crucial in getting the desired performance are:<\/div>\n