Proactive and Attentive Autonomous Navigation and Avoidance of Dynamic and Intermittently-Visible Actors

In the increasingly populated and dynamic world we inhabit, one of the fundamental challenges autonomous mobile robots face is navigating through crowded environments. In addition to creating a more complex environment for robots to traverse, crowds also introduce the challenge of intermittent occlusions - actors in the environment may become temporarily occluded from each others' view by other actors as they move through the environment. Intermittent occlusions can result in scenarios where actors seemingly appear out of nowhere, which can induce erratic or unsafe behavior in a robot's planned trajectory. To mitigate this risk, we propose a novel framework for identifying actors of interest in crowded environments based on observed actor dynamics and predicting their behavior over a short time horizon. These predictions are used to formulate constraints for a model predictive controller, which allows our system to compute an optimal trajectory through crowded environments containing intermittently occluded actors.