In this paper, we propose the reinforcement learning based multi-robot collision avoidance approach by learning collision. Dynamical path re-planning, which is massively used in classical collision avoidance methods, needs overall information of the environment. Also, training agent robots to avoid the collision and pursue a goal point simultaneously is inefficient since the agent should learn two tasks. As the number of tasks that the agent should learn increases, it is difficult to make the performance of an algorithm consistent, which is known as reproducibility issue. To overcome these limitations, Collision Avoidance by Learning Collision (CALC), which learns collision instead of avoiding an obstacle robot is suggested. To solve the collision avoidance problem efficiently, the proposed method divides the problem into training and planning. In the training algorithm, an agent robot learns how to collide with a single obstacle robot and then generates a trained policy. With the trained policy, the agent can pursue a goal point since the policy leads the agent to `collide' with the goal. Furthermore, by taking action in a reverse way from the trained policy, the agent can avoid multiple obstacle robots in the planning algorithm at once. The proposed method is validated both in the robot simulation and real robot experiment, and compared with the existing collision avoidance method.

Distribution-to-distribution normal distributions transform (NDT-D2D) is one of the fast point set registrations. Since the normal distributions transform (NDT) is a set of normal distributions generated by discrete and regular cells, local minima of the objective function is an issue of NDT-D2D. Also, we found that the objective function based on L2distance between distributions has a negative correlation with rotational alignment. To overcome the problems, we present a method using dynamic scaling factors of covariances to improve the accuracy of NDT-D2D. Two scaling factors are defined for the preceding and current NDTs respectively, and they are dynamically varied in each iteration of NDT-D2D. We implemented the proposed method based on conventional NDT-D2D and probabilistic NDT-D2D and compared to the NDT-D2D with fixed scaling factors using KITTI benchmark data set. Also, we experimented estimating odometry with an initial guess as an application of distribution-to-distribution probabilistic NDT (PNDT-D2D) with the proposed method. As a result, the proposed method improves both translational and rotational accuracy of the NDT-D2D and PNDT-D2D.

We consider the problem of assigning reels of components to slots on feeder racks of a surface mounting machine which is used to populate printed circuit boards with surface mount technology. This reel assignment problem (RAP) is one of a series of optimization problems that should be addressed to improve the production rate. We preprocess RAP to facilitate the application of dynamic programming and then formulate it as an integer programming problem. We solve the problem by using a heuristic algorithm based on dynamic programming. We implement the algorithm as a computer program and perform computer simulations. The simulation results are compared to those of the Effective Algorithm previously proposed.

The multiwavelength optical networking (MONET) consortium has demonstrated national-scale optical networking in a multilocation testbed in New Jersey. The demonstration involves transparent optical connections over path lengths as long as 2290 km, through several network elements (NEs) controlled by two interoperating network control and management (NC&M) systems. This paper describes in detail the three constituent testbeds and the experiments.

This paper presents a nonlinear compensation method based on neural networks for trajectory control of robotic manipulators. A multi-layered perceptron neural network (MLP) is used to predict the required actuator torques of a robot to follow a desired trajectory, and these predicted torques are applied to the robot as feedforward compensations in parallel to a linear feedback controller. An acceleration based learning scheme is proposed to adjust the connection weights in the neural network to form an approximated dynamic model of the robot. Simulation results show that the proposed learning scheme improves the speed of error convergence of the system and reduces the convergent error with the efficient adaptation to the changing system dynamics. The validity of the proposed learning scheme is verified through experiments.>