Shin-ichiro Kaneko

Optimal trajectory generation for a prismatic joint biped robot using genetic algorithms

Abstract In this paper, a prismatic joint biped robot trajectory planning method is proposed. The minimum consumed energy is used as a criterion for trajectory generation, by using a real number genetic algorithm as an optimization tool. The minimum torque change cost function and constant vertical position trajectories are used in order to compare the results and verify the effectiveness of this method. The minimum consumed energy walking is stable and the impact of the foot with the ground is very small. Experimental investigations of a prismatic joint biped robot confirmed the predictions concerning the consumed energy and stability.

Control of legged robots during the multi support phase based on the locally defined ZMP

This paper presents a new method for controlling legged robots during the multi support phase, where the robot may have hand contact with the walls, in addition of feet supported on ground. In our method, we consider the contact condition of each foot or hand separately by the locally defined ZMP. The locally defined ZMP are used to plain the robot motion and guaranty the stability. The proposed control algorithm is verified in the real hardware of prismatic joint biped robot. The experimental results show a good performance of the proposed control method.

Evolution of Task Switching Behaviors in Real Mobile Robots

In recent years, much research is focused on evolution or learning of different behaviors. Because both algorithms require much computational time, most of these approaches are conducted in simulated environments. When evolution or learning took place in real robot, a single task was considered. In this work, we evolve neural controllers for task switching behavior using e-puck robots. The e-puck robot has to move to the sound source while first reaching the lights distributed in the environment. Experimental results show a good performance of neural controllers evolved in the real hardware of the e-puck robot.

Pick-place of dynamic objects by robot manipulator based on deep learning and easy user interface teaching systems

Development of autonomous robot manipulator for human-robot assembly tasks is a key component to reach high effectiveness. In such tasks, the robot real-time object recognition is crucial. In addition, the need for simple and safe teaching techniques need to be considered, because: small size robot manipulators’ presence in everyday life environments is increasing requiring non-expert operators to teach the robot; and in small size applications, the operator has to teach several different motions in a short time.,For object recognition, the authors propose a deep belief neural network (DBNN)-based approach. The captured camera image is used as the input of the DBNN. The DBNN extracts the object features in the intermediate layers. In addition, the authors developed three teaching systems which utilize iPhone; haptic; and Kinect devices.,The object recognition by DBNN is robust for real-time applications. The robot picks up the object required by the user and places it in the target location. Three developed teaching systems are easy to use by non-experienced subjects, and they show different performance in terms of time to complete the task and accuracy.,The proposed method can ease the use of robot manipulators helping non-experienced users completing different assembly tasks.,This work applies DBNN for object recognition and three intuitive systems for teaching robot manipulators.

Real Time Robot Policy Adaptation Based on Intelligent Algorithms

In this paper we present a new method for robot real time policy adaptation by combining learning and evolution. The robot adapts the policy as the environment conditions change. In our method, we apply evolutionary computation to find the optimal relation between reinforcement learning parameters and robot performance. The proposed algorithm is evaluated in the simulated environment of the Cyber Rodent (CR) robot, where the robot has to increase its energy level by capturing the active battery packs. The CR robot lives in two environments with different settings that replace each other four times. Results show that evolution can generate an optimal relation between the robot performance and exploration-exploitation of reinforcement learning, enabling the robot to adapt online its strategy as the environment conditions change.

Evolution of low-complexity neural controllers based on multiobjective evolution

In this paper, we present a new method based on multiobjective evolutionary algorithms to evolve low-complexity neural controllers for agents that have to perform multiple tasks simultaneously. In our method, each task and the structure of the neural controller are considered as separated objective functions. We compare the results of two different encoding schemes: (1) connectionist encoding, and (2) node-based encoding. The results show that multiobjective evolution can be successfully applied to generate low-complexity neural controllers. In addition, node-based encoding outperformed connectionist encoding in terms of agent performance and the robustness of the neural controller.