Effective input order of dynamics learning tree
Chyon Hae Kima‡ , Shohei Hamaa§, Ryo Hiraia, Kuniyuki Takahashib , Hiroki Yamadab¶, Tetsuya Ogatab#
and Shigeki Suganob‡‡
aFaculty of Engineering, Department of Electrical Engineering and Computer Science, Iwate University, Morioka-shi, Japan; bWaseda University, Japan
ABSTRACT
In this paper, we discuss about the learning performance of dynamics learning tree (DLT) while mainly focusing on the implementation on robot arms. We propose an input-order-designing method for DLT. DLT has been applied to the modeling of boat, vehicle, and humanoid robot. However, the relationship between the input order and the performance of DLT has not been investigated. In the proposed method, a developer is able to design an effective input order intuitively. The proposed method was validated in the model learning tasks on a simulated robot manipulator, a real robot manipulator, and a simulated vehicle. The first/second manipulator was equipped with flexible arm/finger joints that made uncertainty around the trajectories of manipulated objects. In all of the cases, the proposed method improved the performance of DLT.
KEYWORDS
Learning; modeling; humanoid robot; manipulation; drawing
Modeling the i mpact of i nteraction on pedestrian group motion
Z. Yücela, F. Zanlungob and M. Shiomib
aDepartment of Computer Science, Division of Industrial Innovation Sciences, Graduate School of Natural Science and Technology, Okayama University, Okayama, Japan; bIntelligent Robotics and Communication Laboratories, ATR International, Kyoto, Japan
ABSTRACT
Mobile social robots aimed at interacting with and assisting humans in pedestrian areas need to understand the dynamics of pedestrian social interaction. In this work, we investigate the effect of interaction on pedestrian group motion by defining three motion models to represent (1) interpersonal-distance, (2) relative orientation and (3) absolute difference of velocities; and model them using a dataset of 12000+ pedestrian trajectories recorded in uncontrolled settings. Our contributions include: (i) Demonstrating that interaction has a prominent effect on the empirical distributions of the proposed joint motion attributes, where increasing levels of interaction lead to more regular behavior (ii) Developing analytic motion models of such distributions and reflect the effect of interaction on model parameters (iii) Detecting the social groups in a crowd with almost perfect accuracy utilizing the proposed models, despite the constant flow direction in the environment which causes unrelated pedestrians to move in a correlated way, and thus makes group recognition more difficult (iv) Estimating the level of intensity with considerable rates utilizing the proposed models
KEYWORDS
Personal and service robotics; environmental intelligence; human–robot interaction; pedestrian group motion; social interaction
MPC p olicy learning u sing DNN for human following control w ithout collision
N. Hirose, R. Tajima and K . S ukigara
TOYOTA Central R & D Labs., Inc, N agakute, Japan
ABSTRACT
Model p redictive control h as recently been applied to a wide variety o f motion control s ystems. Model predictive control can be used to generate optimized control inputs with excellent p erformance c o n s i d e r i n g in e q u a l i t y co n s t r a i n t s t o t h e co n t r o l in p u t s , c o n t r o l o u t p u t s , a n d s t a t e v a r i a b l e s . However, the computational load for this method is too h eavy for implementation in most actual s y s t e m s b e c a u s e t h e q u a d r a t i c p r o g r a m m i n g p r o b l e m m u s t b e s o l v e d wi t h i n t h e s a m p l i n g p e r i o d . A s t h e n u m b e r o f in e q u a l i t y co n s t r a i n t s , co n t r o l v a r i a b l e s , an d s t a t e v a r i a b l e s i n t h e co n t r o l s y s t e m increases, more calculation time is required. In this study, a deep neural network d esigned t o learn the model p redictive control policy was developed t o r educe t he computational load. It is expected that a r elatively small n eural n etwork can b e u sed t o learn the model p redictive control policy. In the proposed system, t he motion controller calculates the learned neural network in r eal t ime instead of solving t he quadratic programming problem, r ealizing almost t he same control p erformance as t he original model p redictive control approach. The effectiveness of the p roposed approach was verified by applying it to the control o f a personal r obot designed t o follow t he user, which can p rovide daily support t o t he elderly. In Matlab s imulations, t he calculation t ime for the p roposed approach was approximately 4.5 × 103 times faster than that of the conventional method of solving the quadratic programming problem. In addition, an experiment using an actual personal robot was conducted to confirm the control performance.
KEYWORDS
M o d e l p r e d i c t i v e c o n t r o l ; deep learning; n eural network; quadratic p r o g r a m m i n g p r o b l e m ; o p t i m i z a t i o n
Simultaneous design of parameters and controller of robotic manipulators:
closed loop approach to practical implementation
M. Moradi , M. Naraghi and A. Kamali E.
department of Mechanical engineering, Amirkabir University of technology, tehran, iran
ABSTRACT
In this paper, Simultaneous Design of Plant parameters and Controller (SDPC) of robotic manipulators is presented using the Closed Loop (CL) optimal control approach. Since a robot is inherently used to perform repetitive tasks, the SDPC problem for robotic manipulator systems is practically reformulated as the optimal balancing problem and a demonstration of practical implementation. The optimality conditions are derived using the CL-optimal control theory and then solved by the policy iteration method. The modifid policy iteration is further used to design robot parameters. The results are compared with the Open Loop (OL) solution through the optimal and static balancing cases. As a fial point, although OL is the exact solution, but it has higher control cost over the CL implementation (for the case implemented here the cost diffrence is 16.75%). Also CL-case has
much shorter calculation time (51%) than that of the OL case
KEYWORDS
optimization; simultaneous plant-controller design; optimal control; optimal balancing
