Autonomous Street Crossing with City Navigation Robots

城市自主导航带来了各种各样的难题。其中，最具挑战性的问题之一是自动过马路。虽然过去已经对城市导航车辆过马路的问题进行了深入研究，但最近的大部分工作都集中在穿越有信号的人行横道或需要人类协助的具体任务上。在这个项目中，我们将开发一种完全自主的过马路方法，使机器人即使在没有人行横道或有信号的人行横道的情况下也能过马路。为了实现这一目标，我们将采用基于图像、激光雷达和声音数据的多模式方法来识别交叉机会并跟踪其他交通参与者。此外，我们将采用逆强化学习从人类过马路活动的观察中推断出过马路的策略。

Autonomous urban navigation poses a variety of difficult problems. Among them, one of the most challenging ones is that of autonomously crossing streets. While the problem of crossing streets with city navigating vehicles has been studied intensively in the past, the majority of recent work focused on the specific task of crossing signalized crosswalks or required the assistance of humans. In this project, we will develop a completely autonomous approach for crossing streets that allows a robot to cross a street even when there is no crosswalk nor a signalized pedestrian crossing. To achieve this, we will follow multimodal approaches based on images, lidar and sound data for the identification of crossing opportunities and the tracking of other traffic participants. In addition, we will employ inverse reinforcement learning to infer crossing policies from observations of human street crossing activities.