In this paper, we propose a reinforcement learning-based algorithm for trajectory optimization for constrained dynamical systems. This problem is motivated by the fact that for most robotic systems, the dynamics may not always be known. Generating smooth, dynamically feasible trajectories could be difficult for such systems. Using sampling-based algorithms for motion planning may result in trajectories that are prone to undesirable control jumps. However, they can usually provide a good reference trajectory which a model-free reinforcement learning algorithm can then exploit by limiting the search domain and quickly finding a dynamically smooth trajectory. We use this idea to train a reinforcement learning agent to learn a dynamically smooth trajectory in a curriculum learning setting. Furthermore, for generalization, we parameterize the policies with goal locations, so that the agent can be trained for multiple goals simultaneously. We show result in both simulated environments as well as real experiments, for a 6-DoF manipulator arm operated in position-controlled mode to validate the proposed idea. We compare the proposed ideas against a PID controller which is used to track a designed trajectory in configuration space. Our experiments show that our RL agent trained with a reference path outperformed a model-free PID controller of the type commonly used on many robotic platforms for trajectory tracking.

In this paper, we address an issue on non-existence of a CAD model of an object in a model-based virtual visual servoing system. The approach for reconstructing a 3D model with fewer faces from point cloud data is presented. Because of a number of faces on the model surface, this reconstructed model required less rendering time than a CAD model, consequently, high speed and high accuracy model-based visual servoing can be achieved. We present the experimental results to demonstrate the effectiveness in terms of rendering time, convergence speed, and tracking speed measurements. The comparison of performances of a model-based visual servoing using different models, i.e.- CAD, Coarse mesh, Fine mesh, and Propose models, is also presented.

In this paper, we propose a coordinated motion control algorithm of dual manipulators for handling of an almost rigid object but with non-negligible deformation. By the proposed algorithm, an almost rigid object with non-negligible deformation, such as a cardboard box, a plastic case, etc., is held and manipulated by dual manipulators stably without producing any vibrations. By defining apparent dynamics of dual manipulators for the interface force between the object and its environment and the dissipative dynamics for the internal force applied by the manipulators to the object independently, stable manipulation is achieved. Experimental results illustrate that the vibratory motion of the manipulated object caused by non-negligible deformation is eliminated by the proposed control algorithm while the apparent dynamics of the manipulated object is realized.

In this paper, we propose a vision based compliant motion control method for part assembly work. Some industrial parts are deformed during assembly of parts. If work progress continues, deformation of the part increases, humans check the deformation and adjust force corresponding to the progress state. The proposed method enables a robot manipulator to adjust force applied for assembly work like a human. In our proposed method, force applied for the work is generated by visual information from a camera reading the deformation of the parts. Processing the visual information quantifies the deformation and the data show the work progress. NCC is generally used for template matching, but in this paper we use it for quantifying deformation. Connectors are assembled by a robot manipulator using proposed method and impedance control in experiments. Experimental results are presented to verify the effectiveness of the proposed method.

This paper presents high speed and high accuracy visual servoing system. The algorithm has three major improvements, which can be implemented in practical applications; Firstly high-accuracy pose estimation by using stereo cameras, secondly real time implementation issues with non-real-time image processing platform and thirdly a consideration for industrial position controller. To resolve the issues, position-based visual servoing (PBVS) is adopted and appearance model based virtual visual servoing (VVS) is applied for pose estimation. VVS approach does not compute the stereo matching but directly compares the OpenGL rendered image and camera image for each camera; estimate the position/orientation using VVS independently for each camera; and provides a theoretically optimal compromise among those estimates. To enhance estimation accuracy, a hybrid method of stereo trigonometry for position estimation and weighted least squares for orientation estimation is proposed to combine the information from the stereo cameras. Operation speed is increased by using graphic processing unit (GPU) acceleration and an on-line trajectory generator which can accommodate the variable cycle of the image processing and the fixed cycle of a common robot controller. Finally, some experimental results illustrate the effectiveness of the proposed framework.