Foresighted People Finding and Following

Abstract

Mobile service robots are needed in several applications (e.g., transportation systems, autonomous shopping carts, household activities ... etc). In such scenarios the robot aids the user with tasks that require the robot to move freely across the environment in addition to direct interaction at certain times. Therefore, such a robot needs a strategy to quickly find the user whenever needed, in addition to a strategy that enables the robot to reason about the user's intended destination to be able to follow him in a foresighted manner if the user needs its help at that destination. In this dissertation, we tackle each of those problems separately in a divide and conquer manner.<br /> We present an approach to learn optimal navigation actions for assistance tasks in which the robot aims at efficiently reaching the final navigation goal of a human where service has to be provided. Always following the human at a close distance might hereby result in inefficient trajectories, since people regularly do not move on the shortest path to their destination (e.g., they may move to grab the phone or make a note). Therefore, a service robot should infer the human's intended navigation goal and compute its own motion based on that prediction. We propose to perform a prediction about the human's future movements and use this information in a reinforcement learning framework to generate foresighted navigation actions for the robot. Since frequent occlusions of the human will occur due to obstacles and the robot's constrained field of view, the estimate about the humans's position and the prediction of the next destination are affected by uncertainty. Our approach deals with such situations by explicitly considering occlusions in the reward function such that the robot automatically considers to execute actions to get the human in its field of view. We show in simulated and real-world experiments that our technique leads to significantly shorter paths compared to an approach in which the robot always tries to closely follow the user and, additionally, can handle occlusions.<br /> On the other side, an autonomous robot that directly helps users with certain tasks often first has to quickly find a user, especially when this person moves around frequently. A search method that relies on a greedy approach that do not perform any predictions about the user's most likely location, even when it is provided with background information about the frequently visited destinations of the user, might not be the best option. In this dissertation, we propose to compute the likelihood of the user's observability at each possible location in the environment based on simulations that rely on hidden Markov model based predictions. As the robot needs time to reach the search locations, we take this time into account as well as the visibility constraints. In this way we aim at selecting effective search locations for the robot to find the user as fast as possible. As our experiments in various simulated environments show, our approach leads to significantly shorter search times compared to the greedy approach.