The development of robot systems must be simplified and made less expensive so that they can be used more widely in various industrial fields. We propose a component architecture of a robot-teaching system based on a universal task model, that enables fast and easy development of robot teaching systems for various types of work. This paper demonstrates the development of robot task design and management systems by applying this commponent architectures, and reports the development efficiency verified by simulation and experiments.
A distributed control method for hyper-redundant manipulators is proposed, aiming at applying to complicated and unknown or dynamic environment. A manipulator consists of serially connected joint units, and each unit has the same degrees of freedom as dimension of task space. The proposed method is highly distributed: each unit is controlled by one controller, only neighboring units communicate with each other, and each unit does not have to obtain information from far units. The method is multiple point control: not only the end-effector but also several units can converge to their own desired positions stably. Thus the manipulator can avoid obstacles only if the units which detect the obstacles with their own proximity sensors set their desired positions far from the obstacles. The manipulator makes expansion and contraction motion; it can move into/out of long and narrow space like pipes. All units are controlled so that their displacements may be equal. Hence they have the same and maximum displacement margin. The dynamics of the end-effector is not influenced by the motion of units; that is important for doing tasks with the end-effector. The proposed method is applied to a planar hyper-redundant manipulator with rotational joints, and the effectiveness and usefulness of the method are ascertained by computer simulations and experiments.
This paper presents an approach to developing integrated software for robots in a uniform style of description. First, we introduce a programming style for describing autonomous systems uniformly as platform independent portable modular programs, which is based on BeNet, a real-time parallel computation model, and Java. Then, we propose an approach to developing modifiable modular software for robots in our programming style. In this approach, developers can easily modify and extend an existing information system by adding software modules of two types, Behavior Units (BUs) and Data Holding Objects (DHOs) . Our approach is suitable for building real-time systems for robot behavior control by integrating diverse algorithms for recognition, motion control, planning, learning and communication. By showing how to build an integrated system for a quadruped mobile robot that can communicate with humans, we demonstrate the effectiveness of our approach.
This paper analyzes the motion of a multi-limbed robot in contact with other objects or the ground at multiple points. A limb is defined as a serial link mechanism, such as an arm, a leg, a finger. The motion of the robot is restricted if the multiple contact points are held stationary. In general there are singular postures at which the motion of the robot has an additional degree of freedom. Contact at multiple points also causes the restriction on their sliding directions when they slide. This paper proves that when the robot is at the singular posture, the sliding directions are more restricted. We also show that the closer the robot approaches the singular posture, the more difficult the contact points slide in the directions infeasible at the singular posture. Sliding analysis is important not only for sliding control but also position and force controls of the limbs. We developed a quadruped robot which can take various postures, e.g. lying on the ground and standing on the knees. The motion of the quadruped robot is analyzed as an example of a multi-limbed robot in multi-point contact.
In recent years, a lot of researches have focused on not only transferring human control strategy to robots, but also supporting unskilled men to improve their ability. In general, transferring motor skills from human to human via conventional media is very difficult. However, the physical assistance by the coach is effective for motor skill training and can accelerate learning. A robotic device can be considered as an active-medium which can provide accurate measurement of motion and physical assistance for human. In this paper, the skill transfer by the neural network-based virtual coach using the active-medium is discussed. The task for operators is to keep the virtual inverted pendulum from falling over. The trained neural network can successfully abstract the essential feature of coach and is adequately compensating for unskillful operation of learner. In the training experiments, the skill-up process is investigated and the effectiveness of proposed training method involving haptics is confirmed.
This paper describes a design method of a robot system which consists on mechanically linked plural agents, so as to realize autonomous robots which can adaptively behave in the real world. A difficulty in designing distributed autonomous systems, is how to embed a dynamics for self organizing environment-oriented-action-rules in the systems. So, we propose a robot system which controlled by a decision making method that has oscillator and learning method based on same object functions, and which has a structure of mechanically linked agents that are identically designed, in order that each agent actively interacts with other agents and the outer world. For verifying an usefulness of the system, behavioral acquiring tests in target approaching and obstacle avoidance were implemented by using Distributed Autonomous Swimming Robot that was constructed by the proposed way. Moreover, for learning efficiency in complex tasks such as obstacle avoidance, we proposed Switching Q-learning in which previously obtained action rules were effectively used As a result, the robot acquired simple obstacle avoidance behavior, and time-sequential connections between acquired action rules were observed by interaction-based learning in a distributed autonomous system. That is, it was verified that the proposed system design method was one of the solutions of adaptive systems to complex environments.
The trade-off of exploration and exploitation is present for a learnig method based on the trial and error such as reinforcement learning. We have proposed a reinforcement learning algorism using reward and punishment as repulsive evaluation (2D-RL) . In the algorithm, an appropriate balance between exploration and exploitation can be attained by using interest and utility. In this paper, we applied the 2D-RL to a navigation learning task of mobile robot, and the robot found a better path in real world by 2D-RL than by traditional actor-critic model.
A set of physical feature values is proposed in order to explain impression produced by bodily expression. The concept of designing the value set is based on Laban Movement Analysis which is a famous theory in body movement psychology. Impressions produced by body expression are closely related to these feature values. Each of the feature values is defined mathematically, so that it is easy to implement in a computer. Also the feature values are calculated on general body movements. Therefore, the set can be implemented in body expression systems of robots and computer graphics bodies in order to facilitate quantitative evaluation and forecast of impression produced by their bodies. This paper describes the concept and an example of implementation on a robot. Validity of the set is verified in an experiment.
In this paper, we propose a novel gradient-type iterative algorithm for solving the optimal trajectory planning problems with multiple constraints. Since it is too difficult to generate trajectories considering all constraints at the same time, we introduce the concept of the order of priority into the trajectory generation procedure. In the proposed algorithm, a nominal trajectory is iteratively improved considering the order of priority of the constraints. The procedure is executed based on the gradient function which is synthesized in a hierarchical manner so that the convergence rate of the constraint with lower priority does not change the one with higher priority. The convergence properties of the proposed algorithm are examined. A method to improve the convergence properties when we apply the algorithm to the problem of nonholonomic systems is also proposed. The simulation results of planning trajectories for a 3-DOF planar free-joint manipulator show the effectiveness of the proposed algorithm.
This paper addresses the problem of adaptive tracking control for nonholonomic two actuated wheels mobile robot systems with unknown parameters. An adaptive tracking controller is proposed for a kinematic model of two-wheel mobile robot with unknown kinematic parameters by using integrator Backstepping approach, and the asymptotic stability of the control system is guaranteed by the Lyapunov's direct method. Furthermore, based on a further backstepping approach, an adaptive torque controller is applied to a dynamic model with unknown dynamic parameters. Simulation examples of a mobile robot with two actuated wheels are provided for illustrating the tracking ability of the controller for both the kinematic models and dynamic models.
This study addresses the problem of opening a door by a manipulator mounted on a mobile robot. Generally speaking, the main subject of opening a door is how to prevent the occurrence of internal forces when the robot moves. Many researchers have been reported to overcome the problem based on the force and/or compliant control. These methods performed well, however, the result of the control system became rather complicated. Additionally, they require force sensors to achieve the control. In contrast, this paper presents a new concept to avoid such complications by making elbow and shoulder joints of the manipulator to be passive. It means no control will be performed on these joints. In doing so, the internal forces between the door and the mobile manipulator will vanish with no efforts. In this paper, first we discuss the differences between the traditional methods and the proposed one to describe the idea. Then the performance of the proposed method is investigated through computer simulations.