In this paper, we present a learning platform for robotic grasping in real world, in which actor-critic deep reinforcement learning is employed to directly learn the grasping skill from raw image pixels and rarely observed rewards. This is a challenging task because existing algorithms based on deep reinforcement learning require an extensive number of training data or massive computational cost so that they cannot be affordable in real world settings. To address this problems, the proposed learning platform basically consists of two training phases; a learning phase in simulator and subsequent learning in real world. Here, main processing blocks in the platform are extraction of latent vector based on state representation learning and disentanglement of a raw image, generation of adapted synthetic image using generative adversarial networks, and object detection and arm segmentation for the disentanglement. We demonstrate the effectiveness of this approach in a real environment.

In this paper, an Embedded solution for fast navigation and precise positioning of mobile robots by floor features is introduced. Most of navigation systems tend to require high-performance computing unit and high quality sensor data. They can produce high accuracy navigation systems but have limited application due to their high cost. The introduced navigation system is designed to be a low cost solution for a wide range of applications such as toys, mobile service robots and education. The key design idea of the system is a simple localization approach using line features of the floor and delayed localization strategy using topological map. It differs from typical navigation approaches which usually use Simultaneous Localization and Mapping (SLAM) technique with high latency localization. This navigation system is implemented on single board Raspberry Pi B+ computer which has 1.4 GHz processor and Redone mobile robot which has maximum speed of 1.1 m/s.

The purpose of this study is to develop a motion generation technique based on a double inverted pendulum model (DIPM) that learns and reproduces humanoid robot (or virtual human) motions while keeping its balance in a pattern similar to a human. DIPM consists of a cart and two inverted pendulums, connected in a serial. Although the structure resembles human upper- and lower-body, the balancing motion in DIPM is different from the motion that human does. To do this, we use the motion capture data to obtain the reference motion to keep the balance in the existence of external force. By an optimization technique minimizing the difference between the motion of DIPM and the reference motion, control parameters of the proposed method were learned in advance. The learned control parameters are re-used for the control signal of DIPM as input of linear quadratic regulator that generates a similar motion pattern as the reference. In order to verify this, we use virtual human experiments were conducted to generate the motion that naturally balanced.

This paper proposes a method for motion estimation of consecutive cameras using 3-D straight lines. The motion estimation algorithm uses two non-parallel 3-D line correspondences to quickly establish an initial guess for the relative pose of adjacent frames, which requires less correspondences than that of current approaches requiring three correspondences when using 3-D points or 3-D planes. The estimated motion is further refined by a nonlinear optimization technique with inlier correspondences for higher accuracy. Since there is no dominant line representation in 3-D space, we simulate two line representations, which can be thought as mainly adopted methods in the field, and verify one as the best choice from the simulation results. We also propose a simple but effective 3-D line fitting algorithm considering the fact that the variance arises in the projective directions thus can be reduced to 2-D fitting problem. We provide experimental results of the proposed motion estimation system comparing with state-of-the-art algorithms using an open benchmark dataset.

For natural human-robot interaction, we need to know location and shape of facial feature in real environment. In order to track facial feature robustly, we can use the method combining particle filter and active appearance model. However, processing speed of this method is too slow. In this paper, we propose two ideas to improve efficiency of this method. The first idea is changing the number of particles situationally. And the second idea is switching the prediction model situationally. Experimental results is presented to show that the proposed method is about three times faster than the method combining particle filter and active appearance model, whereas the performance of the proposed method is maintained.

Large workspace and strong grasping force are required when a robot manipulates big and/or heavy objects. In that situation, bimanual manipulation is more useful than unimanual manipulation. However, the control of both hands to manipulate an object requires a more complex model compared to unimanual manipulation. Learning by human demonstration is a useful technique for a robot to learn a model. In this paper, we propose an imitation learning method of bimanual object manipulation by human demonstrations. For robust imitation of bimanual object manipulation, movement trajectories of two hands are encoded as a movement trajectory of the object and a force trajectory to grasp the object. The movement trajectory of the object is modeled by using the framework of dynamic movement primitives, which represent demonstrated movements with a set of goal-directed dynamic equations. The force trajectory to grasp an object is also modeled as a dynamic equation with an adjustable force term. These equations have an adjustable force term, where locally weighted regression and multiple linear regression methods are employed, to imitate complex non-linear movements of human demonstrations. In order to show the effectiveness our proposed method, a movement skill of pick-and-place in simulation environment is shown.

Measuring task complexity of movement skill is an important factor to evaluate a difficulty of learning and/or imitating a task for autonomous robots. Although many complexity-measures are proposed in research areas such as neuroscience, physics, computer science, and biology, there have been little attention on the robotic tasks. To cope with measuring complexity of robotic task, we propose an information-theoretic measure for task complexity of movement skills. By modeling proprioceptive as well as exteroceptive sensor data as multivariate Gaussian distribution, movements of a task can be modeled as probabilistic model. Additionally, complexity of temporal variations is modeled by sampling in time and modeling as individual random variables. To evaluate our proposed complexity measure, several experiments are performed on the real robotic movement tasks.

The loop closure problem is one of the most challenging issues in the vision-based simultaneous localization and mapping community. It requires the robot to recognize a previously visited place from current camera measurements. While the loop closure often relies on visual bag-of-words based on point features in the previous works, however, in this paper we propose a line-based method to solve the loop closure in the corridor environments. We used both the floor line and the anchored vanishing point as the loop closing feature, and a two-step loop closure algorithm was devised to detect a known place and perform the global pose correction. We propose an anchored vanishing point as a novel loop closure feature, as it includes position information and represents the vanishing points in bi-direction. In our system, the accumulated heading error is reduced using an observation of a previously registered anchored vanishing points firstly, and the observation of known floor lines allows for further pose correction. Experimental results show that our method is very efficient in a structured indoor environment as a suitable loop closure solution.

One of the main problems of topological localization in a real indoor environment is variations in the environment caused by dynamic objects and changes in illumination. Another problem arises from the sense of topological localization itself. Thus, a robot must be able to recognize observations at slightly different positions and angles within a certain topological location as identical in terms of topological localization. In this paper, a possible solution to these problems is addressed in the domain of global topological localization for mobile robots, in which environments are represented by their visual appearance. Our approach is formulated on the basis of a probabilistic model called the Bayes filter. Here, marginalization of dynamics in the environment, marginalization of viewpoint changes in a topological location, and fusion of multiple visual features are employed to measure observations reliably, and action-based view transition model and action-associated topological map are used to predict the next state. We performed experiments to demonstrate the validity of our proposed approach among several standard approaches in the field of topological localization. The results clearly demonstrated the value of our approach.

Hierarchical Planning based on Abstraction of World Elements and Operators(HiPAWO) is proposed for mobile robots task planning, where abstraction of world elements is used for hierarchical planning and abstraction of operators is used for hierarchical decomposition of abstracted actions. Especially, a hierarchical domain theory based on JAH(Joint of Action Hierarchy)-graph is proposed to improve efficiency of planning, where a number of same actions are included in both adjacent hierarchical levels of domain theories to provide relationships between adjacent hierarchical levels. To show the validities of our proposed HiPAWO, experimental results are illustrated and will be compared with two other classical planning methods.

In this paper, we are looking for requirements of software, hardware and application for use in network-based robots and also directions in building standardization and research activities by reviewing technical status of the robot industries developing robots. The questions are including awareness of RUPI(Robot Unifies Platform Initiative) activities, target market and applications, hardware specifications, software development technologies, and HRI(Human Robot Interaction). The RUPI committee creates standard and derives implementation software for network-based robots through industrial requirements as like of the results. Many robots have been developed and launched services based on RUPI 1.0 standards. Based on this achievement we are expanding RUPI standard to include thin and thick client robots. The results also show that which one is important and urgent technology in the sense of industrial robotic business.

We propose a planning algorithm to automatically generate a robust behavior plan(RBP)with which mobile robots can achive their task goal from any initial states under dynamically changing environments. For this, task description space(TDS)is formulated, where a redundant task configuration space and simulation model of physical space are employed. Successful task episodes are collected, where A algorithm is employed. Interesting TDS state vectors are extracted, where occurrence frequency is used. Clusters of TDS state vectors are found by using state transition tuples and features of state transition tuples. From these operations, characteristics of successfully performed tasks by a simulator are abstracted and generalized. Then, a robust behavior plan is constructed as an ordered tree structure, where nodes of the tree are represented by attentive TDS state vector of each cluster. The validity of our method is tested by real robot's experimentation for a box-pushing-into-a-goal task.