Foresighted Robot Navigation Using Predicted Human Behavior (P7)

协助人类完成日常任务是机器人技术的一个重要领域。服务机器人可能执行的任务包括支持坐下或站起来，提供行走辅助，从高架板上取下物体或递送物体。每当服务机器人部署在人类环境中时，它们都必须调整自己的行为以适应用户。在这样的环境中规划适当的行动需要跟踪用户，并推断需要机器人支持的地点、时间和任务。因此，为了有效地帮助用户，服务机器人必须能够预测用户预期的导航目标和轨迹。与此同时，机器人需要导航和定位，以便在请求帮助时靠近用户，同时最大限度地减少对用户的干扰。在本提案中计划的工作中，我们的目标是开发一个可靠的预测系统，用于使用观察历史和预测环境特定轨迹的用户运动。此外，我们计划根据预期的用户操作来估计机器人实际需要帮助的地点和时间。基于这些知识，我们将贡献新的技术来产生有远见的机器人行为。我们将了解用户的偏好和需求，实现前瞻性的服务提供，并优化机器人的行为，以实现用户感知的定位和导航。我们将彻底评估开发的技术，并将它们与人类运动预测和用户感知导航的最先进方法进行比较。总而言之，本项目研究的方法将推进用户感知导航领域，并为在人类环境中操作的辅助机器人提供前瞻性服务。

Assisting humans during everyday tasks is an important field of robotics. Possible tasks to be carried out by a service robot include support in sitting down or getting up, providing walking aids, fetching an object from an elevated board, or delivering objects. Whenever service robots are deployed in human environments, they have to adapt their behavior to the users. Planning appropriate actions in such environments requires to keep track of the user and to infer where, when, and for which tasks robot support is needed. Thus, to efficiently assist the user, a service robot must be able to anticipate the user's intended navigation goals and trajectories. At the same time, the robot needs to navigate as well well as position itself in such a way that it is close to the user when help is requested while also minimizing interference with them. In the work planned within this proposal, we aim at developing a reliable prediction system for the user's motions using a history of observations and predicting environment-specific trajectories. Furthermore, we plan to estimate where and when the help of the robot will actually be needed depending on anticipated user actions. Based on this knowledge, we will contribute new techniques to generate foresighted robot behavior. We will learn about user preferences and needs to realize foresighted service offering and optimize the robot's behavior for user-aware positioning and navigation. We will thoroughly evaluate the developed techniques and compare them to state-of-the-art methods for human motion prediction and user-aware navigation. To conclude, the methods investigated in this project, will advance the area of user-aware navigation and foresighted service offering for assistance robots operating in human environments.