International Journal of Humanoid Robotics

In recent years, jointly utilizing local and global features to improve model performance is becoming an important approach for person re-identification. If the relationship between body parts is not considered, it is easy to confuse the identity differentiation of different persons with similar attributes in the corresponding parts. To solve this problem, we propose a feature fusion-based method for person re-identification, which contains three core parts: an adjacency module, a counterfactual attention module and a global difference pooling module. First, an adjacency module is designed to consider the relationship between adjacent body parts and make the features more discriminative. Next, a counterfactual attention module is proposed to conduct counterfactual intervention analysis and encourage the network to learn more useful attention to obtain more fine-grained features. Then, a global difference pooling module is used to learn the global features of a person’s image itself and pay more attention to the important features of the human body. Through the fusion of local features and global features, our model can effectively distinguish the identities of different people with similar attributes in the corresponding parts. Finally, we conduct a large number of experiments and achieve outstanding results on Market-1501, CUHK03 and Msmt17.

Adaptive and cooperative control of arms and fingers for natural object reaching and grasping, without explicit 3D geometric pose information, is observed in humans. In this study, an image-based visual servoing controller, inspired by human grasping behavior, is proposed for an arm-gripper system. A large-scale dataset is constructed using Pybullet simulation, comprising paired images and arm-gripper control signals mimicking expert grasping behavior. Leveraging this dataset, a network is directly trained to derive a control policy that maps images to cooperative grasp control. Subsequently, the learned synergy grasping policy from the network is directly applied to a real robot with the same configuration. Experimental results demonstrate the effectiveness of the algorithm. Videos can be found at https://www.bilibili.com/video/BV1tg4y1b7Qe/.

The inferior sample efficiency of reinforcement learning (RL) and the requirement for high-quality demonstrations in imitation learning (IL) will hinder their application in real-world robots. To address this challenge, a novel self-evolution framework, named task-oriented self-imitation learning (TOSIL), is proposed. To circumvent external demonstrations, the top-K self-generated trajectories are chosen as expert data from both per-episode exploration and long-term return perspectives. Each transition is assigned a guide reward, which is formulated by these trajectories. The guide rewards update as the agent evolves, encouraging good exploration behaviors. This methodology guarantees that the agent explores in the direction relevant to the task, improving sample efficiency and asymptotic performance. The experimental results on locomotion and manipulation tasks indicate that the proposed framework outperforms other state-of-the-art RL methods. Furthermore, the integration of suboptimal trajectories has the potential to improve the sample efficiency while maintaining performance. This is a significant advancement in autonomous skill acquisition for robots.