For robotic grasping tasks with diverse target objects, some deep learning-based methods have achieved state-of-the-art results using direct visual input. In contrast, actor-critic deep reinforcement learning (RL) methods typically perform very poorly when applied to grasp diverse objects, especially when learning from raw images and sparse rewards. To render these RL techniques feasible for vision-based grasping tasks, we used state representation learning (SRL), in which we encode essential information for subsequent use in RL. However, typical representation learning procedures are unsuitable for extracting pertinent information for learning grasping skills owing to the high complexity of visual inputs for representation learning, in which a robot attempts to grasp a target object. We found that the proposed preprocessed input image is the key to capturing effectively a compact representation. This enables deep RL to learn robotic grasping skills from highly varied and diverse visual inputs. Further, we demonstrate the effectiveness of the proposed approach with varying levels of preprocessing in a realistic simulated environment. We also describe how the resulting model can be transferred to a real-world robot and also demonstrate a 68% success rate on real-world grasp attempts.

The use of plastic film substrates for organic electronic devices promises to enable new applications, such as flexible displays and conformal lighting, and a new low-cost paradigm through high-volume roll-to-roll fabrication. Unfortunately, presently available substrates cannot yet deliver this promise because of the challenge in achieving the required combination of optical transparency, impermeability to water and oxygen, mechanical flexibility, high-temperature capability, and chemical resistance. Here, we describe the development and performance of a plastic substrate comprising a high heat polycarbonate film combined with a unique transparent coating package that is aimed at meeting this challenge.

This paper presents task description language (TDL) for underwater robots. General primitives of TDL were designed by considering common primitives of conventional computer languages, and underwater robot-specific command primitives were designed by modifying the turtle graphic commands in Lego. An alias and procedure commands make the TDL intuitive and modular. Concurrent processing and sensor-based event handling commands are also available with TDL. An example of survey task with lawnmower pattern shows feasibility and easiness of TDL.

Presents a neuro-fuzzy controller for autonomous underwater vehicles (AUVs) of which the dynamics are highly nonlinear, coupled, and time-varying. Modified fuzzy membership function-based neural networks (FMFNN) are used to combine advantages of fuzzy logics and neural networks, such as inference capability and adoption of human operators' fuzzy logics, and universal learning capability with neural networks. With initial fuzzy rules given by a human operator or automatically generated by a controller, the AUV can learn appropriate control parameters without human intervention, taking into account the differences in sensor characteristics in different environments. Internal learning loops and simplified derivatives of unknown systems are used for fast and simple converging algorithms. Compared to other control methods, the proposed FMFNN control algorithm never requires any information on systems, off-line learning procedures, or human intervention to adjust parameters. To show the validity of the proposed algorithm, computer simulation results are presented.

A visual servoing algorithm is proposed for a robot with a camera in hand, where fuzzy logics and fuzzy-neural networks are employed to represent and/or learn camera motion commands to track a moving object in terms of image features and their variations. Specifically, novel image features are suggested by employing a viewing model of perspective projection to estimate relative pitching and yawing angles between the object and the camera. And, owing to the uniqueness of the proposed image features, at most two input variables are shown to be sufficient for the design of fuzzy logics and/or fuzzy-neural networks. To compensate dynamic characteristics of the robot, desired feature trajectories for the learning of visually guided line-of-sight robot motion are obtained by measuring features by the camera in hand not in the entire workspace, but on a single linear path along which the robot moves under the control of a commercially provided function of linear motion. And then, control actions of the camera are approximately found by fuzzy neural networks to follow such desired feature trajectories. To show the validity of proposed algorithm, some experimental results are illustrated, where a four axis SCARA robot with a B/W CCD camera is used.

It is shown that there exists a nonlinear mapping which transforms image features and their changes to the desired camera motion without measuring of the relative distance between the camera and the object. This nonlinear mapping can eliminate several difficulties occurring in computing the inverse of the feature Jacobian as in the usual feature-based visual feedback control methods. Instead of analytically deriving the closed form of this mapping, a fuzzy membership function (FMF) based neural network incorporating a fuzzy-neural interpolating network is proposed to approximate the nonlinear mapping, where the structure of the FMF network is similar to that of radial basis function neural network which is known to be very effective in the function approximation. Several FMF networks are trained to be capable of tracking a moving object in the whole workspace along the line of sight. For an effective implementation of the proposed FMF network, an image feature selection process is investigated, and the required fuzzy membership functions are designed. Finally, several numerical examples are presented to show the validity of the proposed visual servoing method.>

A visual servoing method is proposed for eye-in-hand robots. It employs a self-organizing fuzzy controller. There is defined a Jacobian that is not a function of a relative position of the object but is a function of the image features. Instead of obtaining an analytic form of the proposed Jacobian, a self-organizing fuzzy controller is then proposed to alleviate difficulties in real-time implementation. To show its validity, the proposed method is applied to a 2D visual servoing task.>