This paper proposes a human approaching robot navigation framework that enables a mobile service robot to (i) estimate a socially optimal approaching pose, and (ii) navigate safely and socially to the estimated approaching pose. In the first stage, the robot estimates potential approaching poses of a human or a human group, which the robot can safely and socially approach, using the dynamic social zone model. In the second stage, the proposed framework selects a socially optimal approaching pose, then estimate a socially optimal trajectory of the robot using the proposed goal-oriented timed elastic band (GTEB) model. The developed GTEB model takes into account the current robot’s states, robot dynamics, dynamic social zone, regular obstacles and potential approaching poses to generate the socially optimal robot trajectory from the robot’s current pose to the selected optimal approaching pose. The motion control command extracted from the socially optimal trajectory is then utilized to drive the mobile robot to approach the individual humans or human groups, while safely and socially avoiding regular obstacles, human and human groups during the navigation process. The proposed approaching human framework is verified in the both simulation and real robots. The results illustrate that, the mobile robot equipped with our developed GTEB model is able to safely and socially approach and avoid individual humans and human groups, while guaranteeing the comfortable safety for the humans and socially acceptable behaviors for the robot. Note to Practitioners—Although our proposed GTEB model is capable of estimating a socially optimal approaching pose and social robot trajectory, driving the robot to approach a human and a human group, and providing the safety and comfort for humans and socially acceptable behaviors of the robot, there exists a few drawbacks if we wish to apply the proposed approaching human framework in dynamic social environments. First, the optimizer for the GTEB model should be improved in terms of computational time and accuracy to avoid generating unpredictable robot trajectories, especially in dynamic social environments. Second, the socio-spatio-temporal characteristics of the humans including human position, motion and orientation, and human group and human–object interactions play an important role in the proposed GTEB model. However, the existing techniques are only suitable in quasi-dynamic social environments. Hence, highly accurate, robust and real-time algorithms for human detection and tracking, and social interaction detection are necessary. Third, social interactive intentions such as human–robot, human–human and human–object interactive intentions should be predicted and incorporated into the proposed framework to improve the performance of the developed framework in the dynamic social environments. Last but not least, human and human group identification algorithms should be proposed to enable the robot to identify the humans, whose the mobile robot is requested to approach. In the future, the social interactive intentions, the human’s future states and its trajectories and will be predicted using deep learning algorithms and incorporated into the approaching human framework to improve the performance of the proposed framework.
Read full abstract