CONFERENCE: IROS 2023

To realize risk sensitivity and improve the safety of the offline RL agent during deployment, this work proposes a multipolicy control framework that combines offline RL navigation policy with a risk detector and a force-based risk-avoiding policy. In particular, a Lyapunov density model is learned using the latent feature of the offline RL policy and works as a risk detector to switch the control to the risk-avoiding policy when the robot has a tendency to go out of the area supported by the training data. Experimental results showed that the proposed method was able to learn navigation in a crowded scene from the offline trajectory dataset and the risk detector substantially reduces the collision rate of the vanilla offline RL agent while maintaining the navigation efficiency outperforming the state-of-the-art methods.

[video]

CONFERENCE: IAS 2023

In this paper, we propose a socially aware navigation method using reinforcement learning with human action entropy maximization, in which the diversity of human choices is considered as a reward to inform the robot of the socially acceptable interaction with a crowd. By learning to maximize this reward, the proposed method realizes social acceptability by giving pedestrians more choices to enlarge their chance to take their preferred action.

[paper] [slide]

JOURNAL: IEEE Robotics and Automation Letters 2020

Awareness of surrounding and prediction of dangerous situations is essential for autonomous mobile robots, especially during navigation in a human-populated environment. To cope with safety issues, state-of-the-art works have focused on pedestrian detection, tracking, and trajectory prediction. However, only a few studies have been conducted on recognizing some specific types of dangerous behaviors exhibited by pedestrians. Here, we propose a tracking enhanced detection method to recognize people using their smartphones while walking, referred to as smartphone zombie. Features used for pedestrian detection usually involve the rotation variance problem, and in this paper, the drawback is handled by employing motion information from multi-object tracking. The proposed solution has been validated through experiments performed on a newly collected dataset. Results showed that our detector can learn a distinct pattern of the appearance of smartphone zombies. Thus, it can successfully detect them outperforming the existed detection method.

[paper] [video]

-