We present a motion stabilization system for a biped robot that makes it possible to keep relative posture and position to a moving or stationary object. Our system consists of two layers of control subsystems, gaze control system and motion control system. In order to achieve an actual motion which follows exactly a scheduled one, the biped robot gazes a target to estimate errors of robot motion and adjusts both an actual motion and the scheduled one simultaneously. The gaze control system has 2 DOF controller, visual feedback part and feedforward part based on a scheduled robot motion. A periodic motion of robot body swing induced by walking allows us to estimate the distance to the target by forming a motion stereo. The scheduled motion is adjusted based on an adaptive law of Model Reference Adaptive Control (MRAC) .
The software system for the human friendly robot is important to realize a practical flexible robot. I propose the system based on a bilinear time delay neural network model for this purpose. The proposed model can describe both linear and non-linear models for robot software. And this model also can describe both algebraic and differential equation systems. I discuss how this model works for robot functions. I also show the example robot functions of motion, sensing and logic using proposed model.
In this study, we focused on a rehabilitation motion of human wrist joint and aim at developing a mechanical equipment to support these rehabilitation motion. A pneumatic parallel manipulator is introduced since it can drive multiple D.O.F. enough to correspond to a wrist motion and has inherent compliance characteristics resulted from air compressibility works as safety mechanism and minute force regulating function. An impedance control strategy is applied on the manipulator to implement rehabilitation motion by adjusting impedance parameters appropriately. The validities of the proposed multiple D.O.F. rehabilitation system for the several rehabilitation motion are confirmed through experiments.
In this paper, we propose the design guideline of the micro force sensor that becomes important as a force measurement method with a semiconductor strain gauge. The range of the force can be measured is from μN to N, and we controlled the stress applied to the sensor by devising the sensor structure. We developed three types of force sensor to cover such a wide range. The force sensor to measure μN level of force has the cantilever structure, which can transmit the minute force effectively. The force sensor for mN range force is pressure sensor type, and the force can be measured as pressure through the transmission medium to avoid breakage. The force sensor for N range force adopted a new method that the force is directly applied to the diaphragm. We designed each type of sensor based on simulation, and enabled the output voltage range to mV/V regardless of the type. Then we made the prototypes (Maicrogrippar, Catheter Touch Sensor, and Belt Tension Sensor) that applied each sensor. As a result, we verifie the effectiveness of our design guideline for micro sensors to measure several ranges of force.
Multi-legged locomotion enables a mobile robot to behave flexibly in complex environment. High flexible locomotion, however, requires high computational load to the controller. Decentralized control approach is one of the most feasible ways to reduce the load. Although force control is useful in order to realize high adaptability to complex environment, solving a force distribution problem requires generally global information and it does not match the policy of decentralized control system. In this paper, we propose one method to solve a force distribution problem by bottom-up manner, in which local controllers modify the received target values locally by them.