Exploiting Rigid Body Motion for SLAM in Dynamic Environments

The limitations of existing localisation and mapping algorithms in handling highly dynamic environments is a key roadblock in the deployment of autonomous mobile robotic systems in a range of important real world situations. In this paper we propose a technique to integrate the motion of dynamic objects into a Simultaneous Localisation and Mapping (SLAM) algorithm without the need to know a-priori or model the geometry of the object, or even to explicitly estimate the pose of the object. The benefit of this approach lies in a simplification of the underlying SLAM state and a resulting simplification of the non-linear least squares optimisation solution. We demonstrate the performance of the algorithm on two scenarios; SLAM in an urban traffic scenario, and extrinsic calibration of a multi RGBD camera system observing a moving object. Our experiments show consistent improvement in robot localisation and mapping accuracy and demonstrate potential of the proposed algorithm.