Classical map-based navigation methods are commonly used for robot navigation, but they often struggle in crowded environments due to the Frozen Robot Problem (FRP). Deep reinforcement learning-based methods address the FRP problem, however, suffer from the issues of generalization and scalability. To overcome these challenges, we propose a method that uses Collision Probability (CP) to help the robot navigate safely through crowds. The inclusion of CP in the observation space gives the robot a sense of the level of danger of the moving crowd. The robot will navigate through the crowd when it appears safe but will take a detour when the crowd is moving aggressively. By focusing on the most dangerous obstacle, the robot will not be confused when the crowd density is high, ensuring scalability of the model. Our approach was developed using deep reinforcement learning (DRL) and trained using the Gazebo simulator in a non cooperative crowd environment with obstacles moving at randomized speeds and directions. We then evaluated our model on four different crowd-behavior scenarios with varying densities of crowds. The results shown that our method achieved a 100% success rate in all test settings. We compared our approach with a current state-of-the-art DRLbased approach, and our approach has performed significantly better. Importantly, our method is highly generalizable and requires no fine-tuning after being trained once. We further demonstrated the crowd navigation capability of our model in real-world tests.

翻译：经典的地图导航方法通常用于机器人导航，但在拥挤环境中常因“冻结机器人问题”（FRP）而面临挑战。基于深度强化学习的方法虽能解决FRP问题，却存在泛化性和可扩展性不足的缺陷。为克服这些难题，我们提出一种利用碰撞概率（CP）帮助机器人安全穿越人群的方法。将CP纳入观测空间后，机器人能够感知移动人群的危险等级：当环境看似安全时，机器人将穿越人群；而当人群运动剧烈时，则会选择绕行。通过聚焦最具威胁的障碍物，机器人能在高密度人群中避免混乱，从而保证模型的可扩展性。我们采用深度强化学习（DRL）技术，在Gazebo模拟器中针对非协作人群环境（障碍物以随机速度和方向运动）进行训练。随后在四种不同密度的人群行为场景下评估模型性能，结果表明：我们的方法在所有测试场景中均实现了100%的成功率。与当前最先进的DRL方法相比，我们的方法表现显著更优。尤为重要的是，该方法具有高度泛化能力，单次训练后无需微调即可直接应用。我们还在真实世界测试中进一步验证了该模型的人群导航能力。