Robot navigation in crowded environments has recently benefited from advances in deep reinforcement learning (DRL) approaches. However, it still presents a challenge to designing socially compliant robot behavior. Avoiding collisions and the difficulty of predicting human behavior are crucial and challenging tasks while the robot navigates in a congested social environment. To address this issue, this study proposes a dynamic warning zone that creates a circular sector around humans based on the step length and speed of humans. To properly comprehend human behavior and keep a safe distance between the robot and the humans, warning zones are implemented during the robot’s training using deep enforcement learning techniques. In addition, a short-distance goal is established to help the robot efficiently reach the goal through a reward function that penalizes it for going away from the goal and rewards it for advancing towards it. The proposed model is tested on three state-of-the-art methods: collision avoidance with deep reinforcement learning (CADRL) , long short-term memory (LSTM-RL), and social attention with reinforcement learning (SARL). The suggested method is tested in the Gazebo simulator and the real world with a robot operating system (ROS) in three scenarios. The first scenario involves a robot attempting to reach a goal in free space. The second scenario uses static obstacles, and the third involves humans. The experimental results demonstrate that the model performs better than previous methods and leads to safe navigation in an efficient time.

This study presents an approach to autonomous navigation for wheeled robots, combining radar-based dynamic obstacle detection with a BiGRU-based deep reinforcement learning (DRL) framework. Using filtering and tracking algorithms, the proposed system leverages radar sensors to cluster object points and track dynamic obstacles, enhancing precision by reducing noise and fluctuations. A BiGRU-enabled DRL model is introduced, allowing the robot to process sequential environmental data and make informed decisions in dynamic and unpredictable environments, achieving collision-free paths and reaching the goal. Simulation and experimental results validate the proposed method’s efficiency and adaptability, highlighting its potential for real-world applications in dynamic scenarios.