• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1992년도 한국자동제어학술회의논문집(국내학술편); KOEX, Seoul; 19-21 Oct. 1992
• /
• pp.387-392
• /
• 1992

This paper presents a local trajectory generation method which is based on a sequence of reference posture-velocities and the efficient low level control algorithm which constructs the complete smooth curve from the trajectory specification. The reference trajectory generator(RTG) which is in between the local path planner(LPP) and the robot motion controller(RMC) generates a sequence of set-points for each path segments from the LPP and pass it to the RMC. The RMC controls the motions of vehicle which should follow the sequence. In the feedback controller of VMC, the method which compensates robot posture-velocity error correctly is used. These methods are implemented on indoor autonomous vehicle, 'ALIVE' mobile robot. The ALIVE mobile robot system is implemented on the 32bit VME bus system: the two VME CPU's are used for RTG and RMC, while the 80C196KC-based VME board is used for motor controller.

• 대한기계학회논문집A
• /
• 제28권5호
• /
• pp.495-502
• /
• 2004

In this paper, kinematic structure of parallel manipulator having 6-DOF is determined to follow the specific trajectory represented by several curves expressed by the parametric variable functions. In addition, the parallel manipulator is designed to have a high dexterity by considering a kinematic isotropy which can stabilize the motion of the moving platform in the restricted workspace.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2003년도 ICCAS
• /
• pp.2492-2497
• /
• 2003

This paper presents optimal walking trajectory generation for a quadruped robot with genetic-fuzzy algorithm. In order to move a quadruped robot smoothly, both generations of optimal leg trajectory and free walking are required. Generally, making free walking is difficult to realize for a quadruped robot, because the patterned trajectory may interfere in the free walking. In this paper, we suggest the generation method for the leg trajectory satisfied with free walking pattern so as to avoid obstacle and walk smoothly. We generate via points of leg with respect to body motion, and then we use the genetic-fuzzy algorithm to search for the optimal via velocity and acceleration information of legs. All these methods are verified with PC simulation program, and implemented to SERO-V robot.

• Journal of Electrical Engineering and Technology
• /
• 제13권2호
• /
• pp.961-968
• /
• 2018

Recently, researches on automatic recognition of human activities have been actively carried out with the emergence of various intelligent systems. Since a large amount of visual data can be secured through Closed Circuit Television, it is required to recognize human behavior in a dynamic situation rather than a static situation. In this paper, we propose new intelligent human activity recognition model using the trajectory information extracted from the video sequence. The proposed model consists of three steps: segmentation and partitioning of trajectory step, feature extraction step, and behavioral learning step. First, the entire trajectory is fuzzy partitioned according to the motion characteristics, and then temporal features and spatial features are extracted. Using the extracted features, four pedestrian behaviors were modeled by decision tree learning algorithm and performance evaluation was performed. The experiments in this paper were conducted using Caviar data sets. Experimental results show that trajectory provides good activity recognition accuracy by extracting instantaneous property and distinctive regional property.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2012년도 추계학술대회 논문집
• /
• pp.209-210
• /
• 2012

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2011년도 춘계학술대회 논문집
• /
• pp.311-312
• /
• 2011

• 한국멀티미디어학회논문지
• /
• 제21권8호
• /
• pp.969-981
• /
• 2018

Recent advances in machine learning approaches such as deep neural network and reinforcement learning offer significant performance improvements in generating detailed and varied motions in physically simulated virtual environments. The optimization methods are highly attractive because it allows for less understanding of underlying physics or mechanisms even for high-dimensional subtle control problems. In this paper, we propose an efficient learning method for stochastic policy represented as deep neural networks so that agent can generate various energetic motions adaptively to the changes of tasks and states without losing interactivity and robustness. This strategy could be realized by our novel trajectory search method motivated by the trust region policy optimization method. Our value-based trajectory smoothing technique finds stably learnable trajectories without consulting neural network responses directly. This policy is set as a trust region of the artificial neural network, so that it can learn the desired motion quickly.

• 자동차안전학회지
• /
• 제11권3호
• /
• pp.30-36
• /
• 2019

This paper suggests a car-following algorithm for urban environment, with multiple target candidates. Until now, advanced driver assistant systems (ADASs) and self-driving technologies have been researched to cope with diverse possible scenarios. Among them, car-following driving has been formed the groundwork of autonomous vehicle for its integrity and flexibility to other modes such as smart cruise system (SCC) and platooning. Although the field has a rich history, most researches has been focused on the shape of target trajectory, such as the order of interpolated polynomial, in simple single-lane situation. However, to introduce the car-following mode in urban environment, realistic situation should be reflected: multi-lane road, target's unstable driving tendency, obstacles. Therefore, the suggested car-following system includes both in-lane preceding vehicle and other factors such as side-lane targets. The algorithm is comprised of three parts: path candidate generation and optimal trajectory selection. In the first part, initial guesses of desired paths are calculated as polynomial function connecting host vehicle's state and vicinal vehicle's predicted future states. In the second part, final target trajectory is selected using quadratic cost function reflecting safeness, control input efficiency, and initial objective such as velocity. Finally, adjusted path and control input are calculated using model predictive control (MPC). The suggested algorithm's performance is verified using off-line simulation using Matlab; the results shows reasonable car-following motion planning.

• 로봇학회논문지
• /
• 제15권2호
• /
• pp.139-146
• /
• 2020

In this paper, the slope of the footplate is adjusted to compensate for the centrifugal force with a series elastic actuator (SEA) attached to the Segway's body to improve the cornering characteristics during turning. To ensure Segway's driving safety in the curvature motion, it is necessary to compensate for the centripetal force by tilting the footplate to generate inward force from gravity. When the footplate is tilted under the control of SEA, the vertical load on both wheels has been changed accordingly. The frictional force of the wheel has been changed by the change of the vertical force, which requires adjustment of driving torque to keep the curvature trajectory. That is, the driving torque has been controlled to keep the curvature trajectory considering the frictional force caused by the turning motion. Four SEAs are attached to the footplate to control the slope of the footplate and the real curvature motion has been demonstrated to verify the effects of SEAs in the high- speed curvature motion.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2005년도 ICCAS
• /
• pp.113-118
• /
• 2005

Robots today have wider application fields than they ever have before. They need to work close to humans and fluid and compliable motions are expected of them. This requires redundant degrees of freedom for completing specific task. And conventional motion teaching method cannot be applied to redundant link structures. In this paper, the authors present a proficient, cost-effective and intuitive method for motion teaching. New software to apply this method to a humanoid is also presented. This new method utilizes current sensors to determine which joints to rotate. The experiment shown in this paper is a case of closed link where arms cannot move independently due to the restrictions in between the hands. After the input of several passing points of motion trajectory, the curve fitting is performed by the developed software. This software can insert new points, delete erroneous points and modify existing points. The developed motion teaching method is applied to the Kumdo robot, which is developed by the authors.