Localization-aware Trajectory Planning on SE(3) for Intravehicular Robots

Visual-based localization paradigms for autonomous robots depend on a reliable presence and detectability of landmarks in the robot’s surroundings. Without this information, localization will fail, which can result in unpredictable robot behavior as it attempts to re-establish its self-awareness. Localization-aware motion planning attempts to minimize the possibility of unsuccessful feature detection by maximizing the number of features in view. In this paper, a differentiable perception-based objective function is developed to enable an efficient gradient-based optimal control approach for localizationaware optimal trajectory planning in SE(3). The differentiable metric specifically describes the density of features in view of the robot camera, based on a given feature map of the robot’s working environment. The optimal control algorithm is embedded into the RRT*-GBO algorithm to provide a global search capability, thus efficiently handling local minima. Results are presented in simulation for an Astrobee robot traversing the Japanese Experiment Module on the International Space Station, using real sparse maps of the same module provided by NASA.