Tactile sensors of comprehensive functions are urgently needed for the advanced robot to co-exist and co-operate with human beings. Pneumatic tactile sensors based on air bladder possess some noticeable advantages for human-robot interaction application. In this paper, we construct a pneumatic tactile sensor and apply it on the fingertip of robot hand to realize the sensing of force, vibration and slippage via the change of the pressure of the air bladder, and we utilize the sensor to perceive the object’s features such as softness and roughness. The pneumatic tactile sensor has good linearity, repeatability and low hysteresis and both its size and sensing range can be customized by using different material as well as different thicknesses of the air bladder. It is also simple and cheap to fabricate. Therefore, the pneumatic tactile sensor is suitable for the application of co-operative robots and can be widely utilized to improve the performance of service robots. We can apply it to the fingertip of the robot to endow the robotic hand with the ability to co-operate with humans and handle the fragile objects because of the inherent compliance of the air bladder.

To grasp the target object stably and orderly in the object-stacking scenes, it is important for the robot to reason the relationships between objects and obtain intelligent manipulation order for more advanced interaction between the robot and the environment. This paper proposes a novel graph-based visual manipulation relationship reasoning network (GVMRN) that directly outputs object relationships and manipulation order. The GVMRN model first extracts features and detects objects from RGB images, and then adopts graph convolutional network (GCN) to collect contextual information between objects. To improve the efficiency of relation reasoning, a relationship filtering network is built to reduce object pairs before reasoning. The experiments on the Visual Manipulation Relationship Dataset (VMRD) show that our model significantly outperforms previous methods on reasoning object relationships in object-stacking scenes. The GVMRN model is also tested on the images we collected and applied on the robot grasping platform. The results demonstrated the generalization and applicability of our method in real environment.

The goal of reinforcement learning is to enable an agent to learn by using rewards. However, some robotic tasks naturally specify with sparse rewards, and manually shaping reward functions is a difficult project. In this article, we propose a general and model-free approach for reinforcement learning to learn robotic tasks with sparse rewards. First, a variant of Hindsight Experience Replay, Curious and Aggressive Hindsight Experience Replay, is proposed to improve the sample efficiency of reinforcement learning methods and avoid the need for complicated reward engineering. Second, based on Twin Delayed Deep Deterministic policy gradient algorithm, demonstrations are leveraged to overcome the exploration problem and speed up the policy training process. Finally, the action loss is added into the loss function in order to minimize the vibration of output action while maximizing the value of the action. The experiments on simulated robotic tasks are performed with different hyperparameters to verify the effectiveness of our method. Results show that our method can effectively solve the sparse reward problem and obtain a high learning speed.

Intelligent servo control significantly reduces the need to adjust control parameters, and is, therefore, widely used in robot joint control. However, existing intelligent servo control strategies for robot joints have problems of computational redundancy, limited prediction accuracy, and insufficient generalization capability. To solve these problems, this article proposes a servo control strategy for robot joints that is based on the incremental Bayesian fuzzy broad learning system (IBFBLS). First, we construct an intelligent servo control strategy with broad learning system on the basis of fuzzy rules to achieve good self-learning and generalization abilities. Second, the learning parameters of the control strategy are optimized by Bayesian inference to achieve precise joint servo control. Finally, the convergence of the control strategy is enhanced by combining it with Lyapunov theory to constrain the learning parameters of the proposed control strategy. The feasibility and superiority of the proposed control strategy are verified by simulation to compare it with existing intelligent servo control methods. In addition, experiments are conducted using robot joint test bed. Both the simulation and experiments verify that the proposed servo control strategy outperforms other servo control methods with respect to tracking accuracy, stability, and convergence. The root-mean-square error in servo control of robot joints was 0.012 $\%$ , which has been reduced by 55.56 $\%$ compared with the current state of the art.