Sungwoo Yang

This study explores manipulator control using reinforcement learning, specifically targeting anthropomorphic gripper-equipped robots, with the objective of enhancing the robots’ ability to safely exchange diverse objects with humans during human–robot interactions (HRIs). The study integrates an adaptive HRI hand for versatile grasping and incorporates image recognition for efficient object identification and precise coordinate estimation. A tailored reinforcement-learning environment enables the robot to dynamically adapt to diverse scenarios. The effectiveness of this approach is validated through simulations and real-world applications. The HRI hand’s adaptability ensures seamless interactions, while image recognition enhances cognitive capabilities. The reinforcement-learning framework enables the robot to learn and refine skills, demonstrated through successful navigation and manipulation in various scenarios. The transition from simulations to real-world applications affirms the practicality of the proposed system, showcasing its robustness and potential for integration into practical robotic platforms. This study contributes to advancing intelligent and adaptable robotic systems for safe and dynamic HRIs.

As robots become increasingly common in human-populated environments, they must be perceived as social beings and behave socially. People try to preserve their own space during social interactions with others, and this space depends on a variety of factors, such as individual characteristics or their age. In real-world social spaces, there are many different types of people, and robots need to be more sensitive, especially when interacting with vulnerable subjects such as children. However, the current navigation methods do not consider these differences and apply the same avoidance strategies to everyone. Thus, we propose a new navigation framework that considers different social types and defines appropriate personal spaces for each, allowing robots to respect them. To this end, the robot needs to classify people in a real environment into social types and define the personal space for each type as a Gaussian asymmetric function to respect them. The proposed framework is validated through simulations and real-world experiments, demonstrating that the robot can improve the quality of interactions with people by providing each individual with an adaptive personal space. The proposed costmap layer is available on GitHub.

Robot navigation has transitioned from avoiding static obstacles to adopting socially aware navigation strategies for coexisting with humans. Consequently, socially aware navigation in dynamic, human-centric environments has gained prominence in the field of robotics. One of the methods for socially aware navigation, the reinforcement learning technique, has fostered its advancement. However, defining appropriate reward functions, particularly in congested environments, holds a significant challenge. These reward functions, crucial for guiding robot actions, necessitate intricate human-crafted design due to their complex nature and inability to be set automatically. The multitude of manually designed reward functions contains issues such as hyperparameter redundancy, imbalance, and inadequate representation of unique object characteristics. To address these challenges, we introduce a transformable Gaussian reward function (TGRF). The TGRF possesses two main features. First, it reduces the burden of tuning by utilizing a small number of hyperparameters that function independently. Second, it enables the application of various reward functions through its transformability. Consequently, it exhibits high performance and accelerated learning rates within the deep reinforcement learning (DRL) framework. We also validated the performance of TGRF through simulations and experiments.

As the demand for service robots increases, a mobile manipulator robot which can perform various tasks in a dynamic environment attracts great attention. There are some controllers that control mobile platform and manipulator arm simultaneously for efficient performance, but most of them are difficult to apply universally since they are based on only one mobile manipulator model. This lack of versatility can be a big problem because most mobile manipulator robots are made by connecting a mobile platform and manipulator from different companies. To overcome this problem, this paper proposes a simultaneous controller which can be applied not only to one model but also to various types of mobile manipulator robots. The proposed controller has three main characteristics, which are as follows: (1) establishing a pose that motion planning can be carried out in any position, avoiding obstacles and stopping in a stable manner at the target coordinates, (2) preventing the robot from collision with surrounding obstacles while driving, (3) defining a safety area where the manipulator does not hit the obstacles while driving and executing the manipulation accordingly. Our controller is fully compatible with Robot Operating System (ROS) and has been used successfully with three different types of mobile manipulator robots. In addition, we conduct motion planning experiments on five targets, each in two simulation worlds, and two motion planning scenarios using real robots in real-world environments. The result shows a significant improvement in time compared to existing control methods in various types of mobile manipulator and demonstrates that the controller works successfully in the real environment. The proposed controller is available on GitHub.