Trajectory planning for human host tracking and following of slave mobile robot on service-related tasks

Summary form only given. People tracking and following has become an increasing popular topic in recent years. The ability of robots to track the people in the surroundings is essential to many real life applications such as museum guidance, office or library assistance. Another aspect of human-robot interaction is the robot's ability to follow a human target. There are various scenarios where instructions, such as holding books in a library or carrying groceries at a store, are given to the robot when following the target host. Several references that provide more details on this video paper contribution can be found in [1: Chen et al. 2011] and [2: Chen et al. 2011]. The first paper focuses on detecting and tracking moving target in a dynamic environment. The second paper describes the proposed method “Trajectory Optimization” in great length and provides details of performance evaluation regarding robot maneuverability. Some of the important related work includes an architecture of moving object tracking similar to the one in our system being introduced in [3: Wang et al. 2007]. In the DATMO system, the widely used concept of occupancy grid map is adopted to distinguish static and dynamic objects which is similar to the work of [4: Wolf & Sukhatme 2004] Another key aspect of this video paper is regarding motion planning, there are many successful works such as [5: Fox et al. 1997], [6: Ulrich & Borenstein 2000], [7: Minguez & Montano 2004], [8: Seder & Petrovi'c 2007] of auto driving in static environments. However, those methods are designed to reach a fixed goal and assume that the environment and robot states are fully observable. Applying traditional obstacle avoidance algorithms on the target following task directly can fail easily because a moving target can change its speed and moving direction at anytime and the target can be occluded by obstacles. In our work, we propose a motion planner for moving target following. The planner uses an extension of dynamic window approach proposed in [9: Chou et al. 2009] and [10: Chou et al. 2011] to find the collision-free velocities and choose a proper velocity using the A* heuristic search. Proper cost functions are designed for minimizing the distance between the robot and the target and maximizing the possibility that the robot can keep observing the target in a fixed time horizon. Similar to the work of [11: Pomares et al. 2010] which uses a collision avoidance system in the human-robot cooperation, our path planner can also guarantee pedestrian safety and remains collision-free. Additionally, we apply the concept in nearness diagram algorithm proposed in [7: Minguez & Montano 2004] for computing a better estimation of the distance between robot and target and therefore achieve a smooth, non-hesitating performance. This video paper aims to solve tracking and following of a host person in indoor environments using laser range finder on a mobile robot for service tasks. In the past, many researchers have focused on improving robot localization in a complex area, or enhancing robot's ability to detect and track a human target. There are also others who dedicate their attentions on path planning in an uncertain domain as mentioned earlier. However, each of these problems is often tackled and solved independently. This video paper proposes a complete system structure which consists of robot localization, tracking moving target, and path planning under uncertainty, all integrated together to achieve the purpose of human following. In the proposed system, a method called “Trajectory Optimization” is designed to integrate the DWA* navigation and the DATMO system. The proposed algorithm uses heuristic search to find a robot trajectory which can maximize target visibility and minimize the distance between robot and the target simultaneously. Compared to other similar works which often lacks the ability to avoid obstacles while following the target, the proposed method can not only guarantee safe following but also allow the robot the follow the target in the most acceptable manner. In other words, maintaining the maximum target visibility at each time instant. In this video, there are 5 parts demonstrating robot's ability to follow a host target in various environments and conditions. The first part shows the robot follows the target in a set-up environment. The environment is intentionally set up in a way that is more challenging for the robot to follow the target, i.e., narrow gap entrance for the robot to pass through and sudden change in target motions. The system also has the ability to consider whether the robot can navigate safely through the gap (i.e., robot's width). In the second part of the video, the robot must follow the target while avoiding another human who suddenly appear to block the path. Here we demonstrate robot's ability to avoid moving obstacles and still to be able to maintain the target even if it is temporarily occluded. The clip also displays the global map showing the robot and target trajectories as well as DWA* trajectories selected using Trajectory Optimization method. In Part 3, the robot follows a host target in a cafeteria and helps carrying a purchased drink. There are numerous pedestrians appearing in the robot range. Therefore, the robot must be able to track the target accurately and not to be distracted by the surrounding people. In Part 4, another real life environment at a busy computer center is chosen. Here the robot starts following the target and a moving pedestrian suddenly emerges in front and blocks the sight of the target. The robot must stop to avoid colliding with the pedestrian and then continue pursuing the target once the pedestrian is out of sight. Finally, in Part 5 of the video, the robot follows the host target in a library. Not only does the robot have to avoid the pedestrians but it also has to navigate smoothly and accurately in the narrow aisle between shelves. In summary, a complete system structure is developed and a novel following algorithm is proposed to achieve robust human following. The system is unique in a way that the robot can avoid any static or moving obstacles and still able to maximize the target visibility. The system has been tested in set-up environments as well as various real-life scenarios to prove its robustness and efficiency as demonstrated in the video.