In this study, we utilize the mechanism of previous soft bio-inspired robots and propose a soft crawling robot that can move on the ground with various protrusions. The robot has two long flexible arms, which autonomously grasp protrusions and move by pulling and pushing the protrusions. The behavior of the robot is complex; nevertheless, it has only two motors. The complex behavior is realized by interacting with the environment. An actual robot was developed using silicone rubber, experiments were conducted and movements were realized.

In this study, we improved our previous snake-like rescue robot by developing and implementing a stabilizing mechanism. By designing the mechanism to work passively, we achieved a stabilizing function without any CPU and sensors. To demonstrate the effectiveness of the developed mechanism, experiments in which the robot approaches a vertical wall at various angles were conducted. In all cases, the robot could avoid the wall without falling.

Lamarck formed the hypothesis Inheritance of acquired characteristics, which states that offspring inherit characters acquired by their parents. However, this hypothesis has now been rejected in the field of genetics. Baldwin offered an explanation for observations of inheritance of acquired characters, despite the fact that they are not inherited; an idea referred to as the Baldwin effect. To verify the Baldwin effect, many studies have been conducted and various numerical simulations have been demonstrated. However, in conventional simulations, the real world is modeled as a simple environment and various properties of the real world are neglected. As a result, effects of universal properties of the real world in evolution have not been discussed. In this study, we considered that universal properties of the real world play important roles in the Baldwin effect. We used the task of escaping from a hunter, and evolved a fugitive agent in a simulated world in which physical properties are used. We investigated the roles of the real-world properties, through conducted simulations, revealed that a fugitive agent was able to reduce learning time by using the real-world properties, thus supporting the occurrence of the Baldwin effect.

In this paper, we propose the design of a robot with a snake-like body based on a test environment. We explore the abstraction of state-action spaces for reinforcement learning. Additionally, we discuss the versatility of the proposed mechanism by showing that different tasks can be completed by simply changing the reward of the reinforcement learning. Finally, we mention the importance of a body design based on an environment by considering the concept of ecological niches.

Recently, autonomous robots which operate in unpredictable complex environment, for example in the rescue operation, in space development and so on have attracted much attention. Snake-like robot is one of typical example of them. However, to control the robot in the unpredictable environment is very difficult problem, because to model the environment is impossible and much information should be measured to move the robot at the complex unknown environment. In addition, many computational costs are required to calculate control law for many degrees of freedom. In this paper, for the first step, we consider autonomous control of the snake-like robot which moves to a light source on a horizontal plane that has many unknown obstacles. To solve the problems, we show importance of passive mechanism, and realize an autonomous control for a snake-like robot by utilizing the passive mechanism. To demonstrate the effectiveness of the proposed mechanism, we have developed a snake-like robot and experiments have been carried out. As a result effective behaviors have been realized

Acquiring adaptive behaviors of robots automatically is one of the most interesting topics of the evolutionary systems. In previous works, we have developed an adaptive autonomous control method for redundant robots. The QDSEGA is one of the methods that we have proposed for them. The QDSEGA is realized by combining Q-learning and GA, and it can acquire suitable behaviors by adapting a movement of a robot for a task. In this paper, we focus on the adaptability of the QDSEGA and discuss the robustness of the autonomous redundant robot that is controlled by the QDSEGA. To demonstrate the effectiveness of the QDSEGA, simulations of obstacle avoidance by a 10-link manipulator in the changeable environment and locomotion by a 12-legged robot with failures have been carried out, and as a result, adaptive behaviors for each environment and each broken body have emerged.

Reinforcement learning is very effective for robot learning. It is because it does not need prior knowledge and has higher capability of reactive and adaptive behaviors. In our previous works, we proposed new reinforce learning algorithm: "Q-learning with dynamic structuring of exploration space based on genetic algorithm (QDSEGA)". It is designed for complicated systems with large action-state space like a robot with many redundant degrees of freedom. However the application of QDSEGA is restricted to static systems. A snake-like robot has many redundant degrees of freedom and the dynamics of the system are very important to complete the locomotion task. So application of usual reinforcement learning is very difficult. In this paper, we extend layered structure of QDSEGA so that it becomes possible to apply it to real robots that have complexities and dynamics. We apply it to acquisition of locomotion pattern of the snake-like robot and demonstrate the effectiveness and the validity of QDSEGA with the extended layered structure by simulation and experiment.

Reinforcement learning is an adaptive and flexible control method for autonomous system. In our previous works, we had proposed a reinforcement learning algorithm for redundant systems: "Q-learning with dynamic structuring of exploration space based on GA (QDSEGA)", and applied it to multi-agent systems. However previous works of the QDSEGA have been restricted to homogeneous agents. In this paper, we extend our previous works of multi-agent systems, and propose a hybrid autonomous control method for heterogeneous multi-agent systems. To demonstrate the effectiveness of the proposed method, simulations of transportation task by 10 heterogeneous mobile robots have been carried out. As a result effective behaviors have been obtained.

Reinforcement learning has recently been receiving much attention as a learning method for not only toy problems but also complicated systems such as robot systems. It does not need priori knowledge and has higher capability of reactive and adaptive behaviors. However, increasing of action-state space makes it difficult to accomplish the learning process. In most of the previous works, the application of the learning is restricted to simple tasks with a small action-state space. Considering this point, we present a new reinforcement learning algorithm: Q-learning with dynamic structuring of exploration space based on genetic algorithm. The algorithm is applicable to systems with high dimensional action and interior state spaces, for example, a robot with many redundant degrees of freedom. To demonstrate the effectiveness of the proposed algorithm simulations of locomotion patterns for a 12-leged robot were carried out. As the result, an effective behavior was obtained by using our proposed algorithm.

Underactuated systems are those processing fewer actuators than degrees of freedom, but generally they are difficult to control. In this paper we realize the crawling motion by an underactuated robot. By changing the location of the fixed point of the robot to the environment, the constraint of the system is changed. Local accessibility of the system with changing the constraints as a nonholonomic system is shown. The control law based on partial feedback linearization is proposed and the stability of the closed-loop system is studied. We discuss the trajectory planning of the robot based on geometric relation. Experiments were carried out to demonstrate the effectiveness of the proposed controller.

Underactuated systems are those processing fewer actuators than degrees of freedom. Decrease of the number of actuators is a merit from the point of view of size and weight of the system. But generally it is difficult to control the underactuated system. In this paper we realize crawling motion by an underactuated robot. Control law based on partial feedback linearization is proposed and stability of the closed-loop system is discussed on the basis of zero dynamics. To demonstrate the effectiveness of the proposed controller experiments have been carried out.

This paper deals with a neural stabilizing controller of general nonlinear systems. The stabilizing state feedback control law is approximated with a multilayer neural network. Connection weights in the neural controller are determined by a min-max algorithm such that the Lyapunov stability theorem holds via a control Lyapunov function.