A three dimensional dynamic simulation method for a biped robot has been proposed considering collision and friction between the foot and the ground. This method is an extension of the simulation method of open link manipulators and the contact simulation method of rigid body mechanics. This approach enables the investigation of various control algorithms of biped locomotions, and it requires the less cost compared with the experimental investigation method. First, a dynamics of free-falling manipulator is described. And then, a plastic collision model is introduced to realize a constrained motion. Finally, the method is applied to the 20-axis and 15-axis biped robot. The simulation results of motions of walking, jumping, and tumbling down are shown.
It is necessary to consider touch or impact forces when a robot moves in human environments. An actual human's finger with its passive element, i.e. skin and muscle, protects both finger tip itself and outer objects. This paper aims at producing such passive element in a finger joint by employing a movable tension-pulley and a transmission wire, controlling actively the touch torque. It was cleared by the torque experiment that the finger having the passive mechanism controlled actively and softly the touch on the object.
We have proposed a system which enables an operator to feel himself/herself touching an arbitrary surface with edges in virtual environment. This system represents the shape of an object in virtual environment by locally con-structing its shape around his/her finger tip with Shape Approximation Device (SAD), of which each surface has convex and concave edges. In this paper we propose a shape model of an object in virtual environment for this local construction of its shape and an algorythm of shape representation based on this model. We design the shape model so that it is possible to construct the represented shape including curved surfaces and curved edges in real time. In this model an object in virtual environment is divided into quadrangles called patches, and the shape on each patch is described using Bézier surface, which is suitable to describe a complicated shape. We realized a hardware system which could represent a shape which included curved surfaces and curved edges in real time.
It is difficult to design a brain of an autonomous robot. The designer must describe a system that interacts with the dynmaic physical world through sensors and motors. In order to design a desirable behavior, it is necessary to combine a set of modules that interact with each other and with the external world. In this paper, we propose an abstract model for describing such a system, BeNet, a network of processes that change their state asynchronously and periodically. The designer describes modules that interact with the other modules and the environment in parallel, by programming rules for changing the output and internal state and by determining their cycles. Designing on BeNet, you can describe a system completely which interacts with the environment in real time, and in which the role of each module and the flow of information are clear, in a simple manner. We implemented programming environments, BNRB's that enable the designer to program and actualize BeNets for autonomous robots. We discuss on the advantages of our approach to designing a robot brain as a BeNet, showing how to realize brains for desirable autonomous behaviors by integrating parallel modules for perception, motion and decision.
Robot learning such as reinforcement learning generally needs a well-defined state space in order to converge. However, to build such a state space is one of the main issues of the robot learning because of the inter-dependence between state and action spaces, which resembles to the well known“chicken and egg”problem. This paper proposes a method of action-based state space construction for vision-based mobile robots. Basic ideas to cope with the interdependence are that we define a state as a cluster of input vectors from which the robot can reach the goal state or the state already obtained by a sequence of one kind action primitive regardless of its length, and that this sequence is defined as one action. To realize these ideas, we need many data (experiences) of the robot and cluster the input vectors as hyper ellipsoids so that the whole state space is segmented into a state transition map in terms of action from which the optimal action sequence is obtained. To show the validity of the method, we apply it to a soccer robot which tries to shoot a ball into a goal. The simulation and real experiments are shown.
For accurate robotic deburring of an article of cast metal, the analysis and the synthesis of the methodology are discussed. Firstly conventional methods for finishing the cast surface are comprehensively surveyed to point out the essential problems. Based on these analysis, two-mode deburring approach (coarse/fine) are proposed. In the fine mode, based on a linear cutting process model, possibility of the complete finishing by iterative look-and-cut feedback are confirmed and feed rates planning strategy to minimize the deburring time are proposed. The validity of the proposed method has been experimentally confirmed.
A method is proposed that enables deburring robots to accurately determine the surface shape of castings. De-termining the shape around a burr with an estimated error range is essential for accurate finishing. A laser range finder installed on the robot hand in our system provides information on the shape. The data output by the laser range finder is often affected by noise due to the sensing conditions and the property of the surface. We have studied a method that uses median filtering and least-square error fitting to acquire noise-reduced data and a method to estimate the error range of the determined shape statistically. Accurate information can thus be obtained by setting the proper sensing conditions. Application of this methods to an iterative deburring method achieved high accuracy for finished surface.
In this paper, we propose a control method of dynamic passing over steps for a wheeled robot whose center of gravity is high. We are developing a wheeled robot called a ‘variable structure type four-wheeled robot’ which, with a simple structure, has ability to pass over a step. Though this robot has only one degree of freedom for changing the body's structure, it can run on four wheels like a car or on two wheels like a wheeled inverted pendulum. We already proposed a method by combining the transfer from 4- to 2-wheeled mode and the transfer from 2- to 4-wheeled mode to perform an experiment of passing over a step whose height is larger than the radius of the wheel successfully. However, it took about 60 seconds to step up or down in this way because there were four changes of control modes in the procedure. Therefore, we propose a dynamic trajectory control method to make passing over a step speedier. We plan two kinds of dynamic trajectories throughout stepping up. One is the trajectory to realize softlanding at the exchange of the contacting wheel from rear to front. The other is the trajectory considering the collision with the step surface in landing to make stepping smoother and speedier. As the result, we have successfully realized an experiment of stepping up and down by this robot in only two seconds.
Chaos represents one of mysterious rich behaviors of nonlinear dynamical systems. A lot of research efforts have been done for mathematical theory behind chaos. In this paper, we develop a method to provide a mobile robot with the chaotic nature. The chaotic mobile robot implies a mobile robot with a controller that ensures chaotic motion. According to the phase tansitivity, one of the two basic features with the sharp dependence on initial condition that characterize chaos, the chaotic mobile robot is guaranteed to scan the whole connected work space. For scanning motion, the chaotic robot does not require the map of work space. It only requires to measure the normal of boundary when it comes close. A mobile robot with such characteristics may find its applications as a patrol robot or a cleaning robot in a closed floor or building. The sharp dependence on initial condition also yields a favourable nature as a patrol robot since the scanning trajectory becomes highly unpredictable. We design the controller such that the total dynamics of mobile robot is represented by the Arnold equation, which is known to show the behavior of non-compressive perfect fluid. Experimental results illustrate the usefulness of the proposed controller.
This paper describes a motion planning method for dancing robots. The dancing robots entertain many people by their fascinating motions. Because their motions are evaluated based on the feelings of human, the motion planning for them is more difficult than industrial robots. Conventional planning methods can not be applied to the dancing robot. In this paper, we propose a systematic method using the images gotten from the dancing motion performed by human. Firstly, measuring the position of several parts in the target motion performed by human, we construct a refference model using computer graphics. The reference model have enough DOF; it may include unimportant DOF for performing the original motion. Then, the unimportant DOF is selected and restricted by evaluating several proposed indexes. We exmaine the reduced model based on subjective evaluation and show that the original motion can be performed by the reduced model without changing the feeling of human. Finally, to make the motion of the reduced model similar to original one, trajectories of unrestricted joints are modified so as to minimize a model error and the effectiveness is shown based on subjective evaluation.
In order to utilize robots in the service fields, such as in offices and in hospitals, the robots should equip the functions for coexisting with human. In this paper, we propose EBBA (Emergent Behavior Based Architecture) for that purpose. EBBA is constructed by behavior modules. And each module can act concurrently by referring inputs from sensors, its object states and its internal states. And the emergent behavior of the system is defined by a certain set of the object states. Therefore, a planner (one of the behavior modules) can control the emergent behavior of the system by refeering and by changing the other modules' object states and internal states. We have implemented a robot hand system by using EBBA. We also shows that the system can perform tasks efficiently by selecting emergent behaviors as the experimental results.
Feedback control of a manipulator with a passive joint which has neither an actuator nor a holding brake is in-vestigated. The manipulator has three axes in a horizontal plane, with the third joint being passive. The dynamic constraint on the free link is 2nd-order nonholonomic. A trajectory for positioning is composed of simple transla-tional and rotational trajectory segments. The trajectory segments are stabilized by nonlinear feedback, considering the motion of the center of percussion of the free link. Simulations and experimental results show the effectiveness of the feedback control law.
This paper presents a new control method designed to improve drivability on downhill grades. Vehicles normally gain speed when traveling downhill regardless of the driver's desire not to accelerate. Drivers often shift into a lower gear to reduce the vehicle speed. With the control method described here, the shifting operation is executed automatically by using a driver's mental model to infer the intention to decelerate. The driver's mental model is identified by applying “Interactive Dichotomizer 3”to obtain the inductive relationship between vehicle operating data and the driver's intention. An examination was made of the timing difference between automated shifting and manual shifting by the driver. The results indicate that the automatic gear shift performed by the controller should be finished before the driver forms an intention to shift gears. The results of this study verify the effectiveness of factoring the inference of the driver's intentions into powertrain control systems.