Research topics

Annual Reports of Research Activities: 2002, 2003, 2004

A dynamic motion analysis platform is constructed to study the nature of human movement. This system consists of a 3D motion capture system which has 6 high-speed cameras, 8 force plates that can measure 3 dimensional foot forces, 32ch EMG sensors and 10 3-axis acceleration sensors. The motion capture system performs online the calculation of body motion from the camera inputs and sends the data to the 3D virtual human CG for real time mimic of human motions. Unlike usual systems that approximates the human motion forces from the numerical differentiation of the calculated angle information, this platform can obtain exact human motion control force by integrating acceleration sensor information with all other inputs, which makes it possible for deeper understanding of human dynamic motion control functions and is thus useful for bio-mimetic robots.

In this research, we construct a novel simulation environment for developing human interaction robot. This system uses AT-clone computers to calculate 3-dimensional dynamics and collision of a robot when it interacts with human and projects the audiovisual results using an immersion type display. Through dynamic calculation of the robot, human can interact with it directly in real time. This system makes it possible for us to design and examine the next generation of human interactive robot easily and safely. Experiments as a virtual robot interacts with human and performs dynamic motion show the effectiveness of our system.

Unlike the present manipulator control technologies that operate objects only by the robots' end-effectors, human beings can easily perform whole body manipulation flexibly and safely. Bio-mimetic research of such skillful human motor behavior is important not only for deeper understanding of human sensations and nervous control functions but also for developing of higher level robots. In this study, we investigate modeling and control of multi-points whole body interaction between a robot manipulator and its object. As an example, we treat a 2-link planar Whole Arm Manipulation (WAM) system and Whole Body Manipulation (WBM) system that operates a circle object with dynamic and static arm frictions. Reasonable dynamic tasks and stable postures during object's operation are investigated from the optimal torque distribution point of view. Cooperative manipulation of a more complex multi-linked object is also studied.

An autonomous decentralized network control system of a robot is constructed via internet. This robot has totally 49 D.O.F. and can communicate with human to perform the required task based on its sound recognition and vision information. In order to realize real time control of so many D.O.F., 9 sets of AT-class PC are involved in controlling each module, such as a head, two arms and two multi-fingered hands, sound recognition and conversation as well as visual recognition. A server is used to coordinate the discrete events between each module. Bio-mimetic researches on hand preshaping when approaching to a specific object, as well as body languages with respect to the sound conversations are introduced into the system so as to realize the human friendly interaction.

An immersion-type tele-interaction system is constructed, which can realize the effect such that the human operator feels as if he/she is directly within the body of the virtual/or real robot and perform the real tasks via the real robots. This technology has greatly extended the task ability as well as performance of tele-manipulation. The developed system has three levels of applications, they are: (1) The virtual robot along can be used easily to test various bio-mimetic control algorithms, such as learning cognitive behaviors.(2)The simulator together with human interface can be used for training the human operator to perform some cooperative tasks in complex environment. (3)The overall system can be used to examine the cooperative tele-manipulation including the physical environmental interactions.

A linear artificial muscle actuator using ionic polymer-metal composites (IPMC), which is an electro-active polymer that bends in response to low voltage (2 or 3 Volt) of electric stimuli, is developing. This actuator owns many excellent properties as biological muscle such as soft and silent. One of the main objectives here is to apply it to the bio-mimetic robots especially the biped robots since the back drive ability is important for efficient biped walking. Within this linear actuator, the elementary component consists of four IPMC films, two are connected by a flexible conductive material. Such elementary units can be connected with each in serial and parallel easily without electric short and any spatial interference.

Human performs beautiful biped walking and can walk a long distance with only small energy supply. This is because we effectively use the gravity based on our body's physical dynamics. A robot should also walk naturally based on its physical dynamics effectively. Such a consideration is referred to as "dynamics based control". McGeer's passive dynamic walking has been considered as a clue to elucidate for natural and energy-effective walking. We analyze the mechanisms of passive dynamic walking and apply it to gait generation for actively controlled walkers. Throughout this study, we suggest that, the dynamic bipedal walking can't be realized by only applying the existing control theory but depends mainly on the deep understanding of the complex physical characteristics of the mechanical system.

Biological systems have fascinating abilities to process complex spatial-temporal information within real time and to recognize their changing environment via vision. Bio-mimetic study of vision system will lead to dramatic improvement of robot's environmental adaptability. By mimicking the neural network of retina, the silicon retina can realize various useful spatial-temporal filtering functions. Based on these filtered outputs, simple algorithm is proposed to calculate optical flow of the moving object in the complex image ground. In addition, by control the parameters of the silicon retina, the vision system can adapt to the changing lighting conditions and track the moving object. FPGA is successfully applied, which greatly reduced the vision system's processing time.

A light and small-size general robot controller is developed and is succeeded applied to control a quadruped robot. This controller is constructed in the form of a hierarchical modular structure. The complex information processing of the robot's environment and the corresponding robot's motion planning is performed at the upper layer of AT-class PC network. The low layer controller, with its size 9cm x 6cm, communicates between the upper layer and the lower level controller to realize the real time management of the robot motion, and the lower level modules (4cm x 3cm) have multi-channels of A/D, D/A and counters interfaces and can easily interact with the actuators and sensors of the robot.

It is well known that the adaptive harmonic gait patterns are mainly generated by the local interaction between nervous oscillators in lower nervous system. A multi-legged robot, named "Caterpillar", has been developed which uses a hierarchical structure inspired by the animal's control structure. The upper layer makes plan for the movement of the whole body, expressed by weighted sum of three primitive motions, and broadcasts the weight to the lower layer. The lower layer consists of six sub-systems and has a modular structure. Each subsystem controls autonomously the swing timing of its leg through the interaction between neighbors and decides the leg trajectory according to the weights from upper layer and its arranged position in the robot. Currently, the plan for constructing a new robot, named MoNOLeg, is in progress.