This paper describes a walking control scheme for a quadruped robot. In conventional walking control scheme, the leg motion is generated from the robot moving trajectory beforehand, and it is difficult to control in real time. The proposed control scheme employs a real time command from a human operator. The operator gives a current planar moving velocity command to the robot. Since the command in the future cannot be predicted, the scheme does not generate a periodic gait. Body propulsive action is basically continued according to the command unless a problem arises. When a foot reaches the border of the work space of the leg, the body propulsive action is interrupted, and a supporting foot pattern is changed according to the leg transition rule. That is, a foot arrival at the border is considered as an event to trigger an recovery action for the body propulsion. The actions are decided to converge to the intermittent crawl gait when straight walking command is continued. The proposed control scheme is evaluated using actual walking robot.
The brain of an autonomous robot generates signals to the motors in each situation of the dynamical real world to achieve various tasks. It will be a complex system that maps the sensation and the intention of the robot into the commands to the actuators, when the robot makes many actions and has many degrees of freedom. In this paper, we propose to describe a hierarchical and modular motion system as a BeNet, an asynchronous network of real-time computational modules that represents a dynamical system completely, which interacts with the physical world. The designer decomposes the system into parallel modules which have distinct roles, and builds an understandable and desirable system, based on our abstract computational model. We show how to design a hierarchical BeNet for desirable actions of an autonomous legged robot, utilizing our programming environment. The system consists of pattern generators and action generators whose roles are stipulated so that the system is easy to debug or extend. In our environment, we can describe a modular system and verify it immediately with a real robot. Each module is described briefly in C and desirable actions of a robot with multiple DOF's were generated by the system.
This paper describes a method of controlling a gaze of a mobile robot for interpretation of road intersection scenes considering the uncertainty of interpretation. From a monocular color image, candidate regions are extracted for possible objects such as curve mirror. The probabilities of the regions coming from the objects are calculated from the probabilistic models of the objects. From the probabilities of the regions and the relation between the objects and intersection types, the current probability distribution of intersection types is calculated. If the entropy of the probability distribution is low enough, the robot adopts the best hypothesis. Otherwise, the robot selects and gazes the part which minimizes the expectation of the entropy. These actions are iterated until the entropy becomes low enough. The experimental results are shown for actual intersection scenes.
The purpose of this research is to propose and develop a motion control method which has adaptive and emergent ability such as biological system. In developing a motion controller, it becomes a hindrance that parameters of robots undergo changes as time passes. Muscular systems have much nonlinear properties and those parameters often change, however, motion controller of biological system can handle them properly. It is to be desired that motion pattern should be changed according to motion parameter changes. So, we try to develop these types of motion controller by use of RNN (Recurrent Neural Network) and GA (Genetic Algorithm) . Parameters in the RNN are treated as the genetic information. Time constants, synaptic weights and thresholds of neurons are crossed over and mutated and suitable motion patterns would be obtained according to the performance indices. We tried to learn and control locomotion of biped robot which parameters are unknown. As a result of simulations, the motion controller acquired several patterns of gaits. Therefore, it is developed that, the motion controller which can learn and generate motion patterns as to parameters of motion system and changes of environments.
We report on our preliminary studies of a new controller for a two-link brachiatiog robot. Motivated by the pendulum-like motion of an ape's brachiation, we encode this task as the output of a“target dynamical system.”Numerical simulations indicate that the resulting controller solves a number of brachiation problems that we term the“ladder”, “swing up”and“rope”problems. Preliminary analysis provides some explanation for this success. We discuss a number of formal questions whose answers will be required to gain a full understanding of the strengths and weaknesses of this approach.
To realize an actuator with flexibility, it is effective to use pneumatic actuators because of pneumatic compressibility. In this paper, a new type of pneumatic rubber actuator is developed. The actuator is called a hexahedron rubber actuator (HRA) . HRA has a hexahedron structure constructed by four plates and two fiber nets, which cover a rubber bladder. Since a HRA produces a rotational motion, it is suitable for making a rotational joint without adding mechanical parts . In this paper, we reveal the static characteristics of the HRA. Furthermore it is experimentally demonstrated that the HRA can realize a good position control performance without oscillation by using hierarchical feedback control. Moreover, we propose an adaptive gain method for the position control of the HRA to simplify the gain tuning. The effectiveness of the proposed method is confirmed through some experimental results.
In industrial assembly lines like for automobile manufacturing, seam sealing is an important process where sealant is painted on seams locate at the joints by thin plate parts for watertight or a rust prevention. Manual tasks in sealing is so exhausted even for skillful human operators because the curve to be sealed is too complex to trace in high-speed. Therefore, most of the sealing processes have already been automated by teaching-playback robots. However, it requires much redundant amount of sealant to cope with the positioning error or shape deformation of the workpiece. This is one of the problems degrading the manufacturing cost. In this paper, we propose a highly reliable algorithm for seam position computation from sensed profile range data adjacent to the seam. The data is assumed to be sensed by a high-speed scanning laser range sensor that is equipped with a robot arm with the sealing gun. Our algorithm employs statistical jump detection from the range data because the jump in the data is a robust feature and the computation for the detection is fairly small. Based on the computed seam position, the sealing gun can be controlled as tracking the seam at such high speed as 300 [mm/sec] . It is proved experimentally that the sealing system with the developed algorithm is very effective, especially in the point of reduction of the redundant sealant.
This paper presents how the control performance of the feature-based visual servo system is improved by utilizing the redundant features. It also proves the effectiveness of the redundant features to increase the accuracy of the visual servo system by using the smallest singular value of a matrix called image Jacobian. Real time experiments on a PUMA 560 manipulator show usefulness of the redundant features to improve the performance.
In this paper, we propose to develop a modular real-time robot vision system in a programming environment based on BeNet. BeNet is a parallel computational model for describing real-time systems completely and briefly by software as a set of computational modules of which roles are distinct. The real-time behavior of BeNet does not depends on the hardware and is easy to predict because each module changes the state and the output at regular intervals. An environment in which one can describe a BeNet and implement it immediately on a multi-processor system with image frame memories, makes it possible to develop a large-scale modular system by integrating diverse methods for computer vision in a short term. We present how to design and develop a system that can find an object and track it in real-time based on the top down input of the attention in our environment. We realized a network of modules for feature extraction, object extraction and attention control, and confirmed that the BeNet shows a desirable behavior in the real world and that it is applicable to an autonomous robot.
In this paper we address the problem of reconfiguration of a freely floating planar space robot. Such a system is nonholonomic in nature due to the conservation of its angular momentum. This paper presents a smooth and time-invariant feedback control strategy that asymptotically converges the system states from practically any configuration to the desired configuration. The controller does not render the desired configuration asymptotically stable in the sense of Lyapunov but suffers from no convergence problem. The control strategy, though time-invariant, uses a nonlinear oscillator and extends the concept of geometric phase to control. In certain situations the controller has a slow rate of convergence but this problem can be easily rectified by simple modifications, as suggested in this paper. A stability analysis of the closed loop system using the original controller is only presented but results of numerical simulation indicate that both the modified controllers as well as the original controller can converge the system states to their desired values satisfactorily.
Manipulators with free joints fall in a class of second-order nonholonomic systems whose dynamical constraints are nonintegrable. Control of such systems is one of major research topics in robotics and control engineering. In the literature, the authors analyzed nonlinear behaviors of a two joint manipulator with a free joint in response to a periodic input. The manipulator showed chaotic behaviors for larger input-amplitudes. A control method to position both joints via modulation of the input-amplitude was also proposed. Although the strategy was effective, it sufferes from heuristics and difficulty for generalization. In this paper, behaviors of the free-joint manipulator with a periodic input are approximated via the 1st-order and 2nd-order averaging methods. The invariant manifolds of the averaged behavior are analyzed and identified by defining a “pseudo-Hamiltonian” on an “input-amplitude normalized phase plane”. A control method to reach a desired invariant manifold via modulation of the input-amplitude is also proposed and a termination control at the destination is formulated. The effectiveness of the proposed control methods is verified by computer simulations.
There are positional and geometric uncertainties in assembly operation. In order to perform an assembly operation by robot successfully, it is necessary to estimate the correct position and geometric information of a fixed environment. Authors have given a technique that can estimate the contact position and geometric information between the object and the environment from a number of positions and orientations of a moving object, which are measured while the object is dexterously probed on the environment around the contact position. This paper considers proposing a method for generating the probing operations automatically. Firstly, the realizability of probing operation on a contact position is discussed. Then, automatic generation approaches for probing operations are described. Lastly, numerical example verification on the proposed approaches is performed and its results are outlined.