Manipulation laboratory at Meiji University investigates many robotic systems such as robotic hands, elastic joint robots, bipedal robots, robot arms, passivity-based control, etc. Especially, as in the laboratory title, we are comprehensively working on the manipulation of robotic hand systems.
As shown in the figure below, robotic hands' research mainly consists of three parts: mechanical design of robotic hands, motion (grasping) controller design, and decision making. To design robotic hands, we primarily use tendon-driven mechanisms that enable us to create small robotic systems. To dynamically control the grasping, we are mainly using the passivity-based control methods. At the current stage, it is not easy to realize autonomous decision-making for grasps. Thus, we use supervisory control systems, where the humans determine operation, while the robotic hand systems automatically grasp and manipulate an object using the passivity-based controllers. We are integrating these methods to realize the practical robotic hand systems.
Tendon-driven mechanisms (TDMs) are the transmission to drive linkages in robotic systems using traction force of wires. The figure below left shows a typical example of the TDM. The circles in the left and right express the motors and a pulley, respectively, and the lines between them express the wires. The motors pull the cables, and then the linkage moves.
TDMs sometimes refer as "wire-driven mechanisms" and "cable-driven mechanisms." However, the TDMs do not perfectly match these mechanisms because cable and wire are named for the hardware but might generate bilateral forces. In contrast, TDMs can generate only unilateral traction forces, which is similar to the musculoskeletal system.
TDMs have several advantages to the conventional robotic systems: First, TDMs can eliminate actuators from linkages for decreasing the weight of the link. Therefore, the TDMs are frequently used for small robotic systems such as robotic hands. Second, some elastic elements can be easily inserted into the TDMs to realize safe human-robot interactions,
as shown in the figure below right. For example, when a robot requires synergetic motions and soft and safety contacts rather than accurate position control, we can replace the actuators at the tendons' ends with springs. Then, the TDMs can adaptively contact the environments and do not hurt them. Third, the TDMs can asymmetrically distribute the actuators to efficiently assign the driving force, as shown in the human's shoulder.
In contrast, the control system of the TDMs is more complicated than the conventional motor drive systems because TDMs require more tendons than joints. Besides, it is challenging to design the TDMs. Therefore, we are also investigating the control systems and the design methods of the TDMs. The following figure shows the classification of TDMs. TDMs can be categorized into the six subclasses according to the existence of the joint equilibrium and the driving DOF. Based on this classification, we have developed a design method for TDMs (Ozawa et al., IEEE TRO 2014).
The photo bottom left shows the designed robotic hand using TDMs, based on the developed design method. The hand consists of the thumb, index, and middle fingers(Ozawa et al., Autonomous Robots, 2014). The fingers' tip two joints connectedly move in free space while adapting to the environment in the constraint situation. The photo bottom right shows the prosthetic hand that uses only three actuators to realize four different grasps. The size and weight of the hand are about 200mm and 440g, respectively, which corresponds the adult hands.
R. Ozawa, K. Hashirii, H. Kobayashi, "Design and Control of underactuated tendon-driven mechanisms",
Proc. of the 2009 IEEE Int. Conf. on Robotics and Automation, Kobe, JAPAN, pp.1522-1527, May, 2009
Conventional dynamic manipulation methods of robotic hands require a lot of information, which includes mass, inertia, shape and position of COM of a grasped object, contact points and so on. This makes it difficult to manipulate an object in an unstructured environment and to introduce robotic fingers in our daily life.
We have developed a dynamic manipulation method for robotic fingers without object information. This controller was inspired by the opposed motion in the precision grasp. The demonstration can be confirmed in the following movies.
References & Selected publications:
S. Arimoto, P. Nguyen, H. Y. Han, and Z. Doulgeri"
Dynamics and control of a set of dual fingers with soft tips "
Robotica, vol. 18, no. 1,pp. 71-80, 2000
Teleoperated systems for robotic hands are usually controlled with master-slave controllers. The master-slave controller is a kinematic mapping from some quantities of human hands, which are joint angle, fingertip position and sometimes pose, to the robotic ones. The kinematic structure of both human hands and robotic hands are complex and some kinematic inconsistency between them exists. Thus, the operation may not work well even if robotic hands perfectly tracks human hands, because the perfect kinematic mapping does not
guarantee the stability of manipulation. Therefore, we construct a semi-automatic teleoperated system for robotic hands.
In our method, the feedback loops of the slave controllers are closed at the slave side, and an operator only sends some scalar values to the slave controllers. Thus, it is easy to execute in a real system, even there is a time delay in communication channels. We have applied the proposed controller to a 2DOF planar hands and a three-ringered robotic hand.
In the systems, we used 2DOF special mechanisms and datagloves as master devices.
The demonstration can be confirmed the following movies.
Selected publications:
Y. Yoshimura, R. Ozawa, "
A supervisory control system for a multi-fingered robotic hand using datagloves and a haptic device",
Proc. of the IEEE Int. Conf. on Robots and Systems (IROS), Vilamoura, Portugal, Dec. 11, pp. 5414-5419, 2012
R. Ozawa, N. Ueda,"
Supervisory control of a multi-fingered robotic hand with data glove",
Proc. of the IEEE Int. Conf. On Intelligent Robots and Systems (IROS), Oct 29-Nov.2,
San Diego, USA, pp.1606-1611, 2007
R. Ozawa, T. Yoshinari, H. Hashiguchi, S. Arimoto,
"Supervisory Control Strategies in a Multi-Fingered Robotic Hand System",
Proc. of the IEEE Int. Conf. on Intelligent Robots and Systems (IROS), Oct 9-15,
Beijing, China, pp.965-970, 2006
An image-based visual servoing(IBVS) is one of the representative method to stablize the robot motion using video image. In this research, we propose a IBVS for controlling the position and orientation of an unmanned aerial vehicle (UAV) using a fixed downward camera observing landmarks on the level ground.
In the proposed method, the visual servoing of the image moments is used to control
the vertical motion and rotation around the roll axis. In contrast, an undesired positive feedback araises in visual errors, as shown in Pheases 1 and 2 of the top left figure, because of the under-actuation of the UAV and this positive feedback makes it difficult to apply the visual servoing to the horizontal motion. Thus, a novel control method using the virtual spring is introduced to control the horizontal motion, as shown in the top right figure. This control method can stabilize the position and orientation of the UAV, as shown in the middle figures. The image during feedback is shown in the bottom figure.
Many researchers have been working about walking and balance control of bipedal robots for the last four decades and have accomplished brilliant results on the fields. Almost all these researches are executed by using model-based control method (MBC) that requires detail models on robots and environment and observation of the zero moment point (ZMP). The ZMP is the point where the moment causing a fall is zero and can be observed by using the force sensors. These requirements decreses the robustness to unmodelled dynamics or force sensor noizes. In general, modeling of fullbody motion is very complex. Thus, in MBC, the biped is usually modelled as a lumped mass and the fullbody motion is separately controlled while the lumped mass balance is carefully preserved.
Thus, we are trying to control the biped robot using the different strategies called passivity-based control methods (PBC). In general, PBC can control robotic systems using less information and is more robust to the modeling errors than MBC. However, it is well known that the controller design is more difficult than MBC because of the stability analysis. Therefore, this method is merely applied to the bipedal control.
We have already developed passivity-based balance control methods for a bipedal robots in multi-contact cases. We have experimentally confirmed that the method worked well for controlling the balance and fullbody motions simultaneously and is robust to the model errors on the environments.
When the endpoint of a robot is controlled, the robot needs to observe the endpoint using the combination of the internal sensor and the kinematic information of the robot or the visual sensor. The former stably provides the endpoint information but the accuracy depends on the accuracy of the kinematic information. The latter is robust to the kinematic errors
but the occlusion might destroy the stability of the controller.
In this research, we have developed a controller that combines the merits of both the endpoint estimation methods. The visual sensor is used to realize the accurate control when the endpoint can be observed while the kinematic information and the internal sensor are used to stably control the endpoint by the sacrifies of the accuracy in the occluded case.
The visual sensor sometimes failed to get the endpoint information but the control worked stably.
Selected publications:
R. Ozawa, Y. Oobayashi, "Adaptive task space PD control via implicit use of visual information", Proc. of the Inter. Symp. on Robot Control," Gifu, Japan, pp.209-215, Sep. 2009
When robots do some tasks in our daily life, they need to control the interaction forces among the robots, human and the environment. Traditionally, this control was actively executed using the feedback from force sensors. In contrast, since the early 1990,
researchers have worked about the hardware compliance to improve the response and the energy efficiency.
We have developed many control methods for the robots with hardware compliance and have considered the relationship between the hardware compliance and the controllers to efficiently use the hardware compliance.
explanation of the video The hardware compliance of the upper robot is controlled using the proposed method while the bottom uses the constant spring. The proposed method adaptively control the optimal impedance to save the energy without knowing the endpoint mass. Thus, the actuator, the left box, stops while the endpoint keeps moving. This means that the robot perfectly uses the kinetic energy preserving in the spring for the endpoint motion.
R. Ozawa, H. Kobayashi:
"Stability of PD Control Systems of Tendon-Driven
Mechanisms with Nonlinear Tendon Elasticity", Proc. of 2000 Japan-USA Flexible
Automation, Ann Aber, Michigan, July, 2000
R. Ozawa, H. Kobayashi:
"Control of Coupled Tendon-Driven Mechanisms with
Nolinear Tendon Elasticity", Proc. Pioneering Int. Symp. on Motion and Vibration
Control in Mechatronics, Tokyo, Japan, April, pp.151-156, 1999
Conventional mass measurement systems such as a bath scale, a libra and so on use "GRAVITY" to weigh. Thus, we cannot use them in a micro - gravity environment such the outer space. In this case, for example, a sample is attached at the tip of an elastic beam, and then, the period of the beam is measured to estimate the sample's weight from the natural frequency when the beam is vibrated by an external force. This method is not robust because the period is easily perturbed by disturbances.
Our developed method is to control the stiffness of the blade spring to converge the natural frequency to a desired one. The controller, which adjusts the effective length of the blade spring, is a simple integrating controller and absorbs disturbances. Thus, it can estimate sample's mass more accurately.
The left figure shows the experimental setup. The actuator unit in the below vibrates the whole upper part and the linear actuator on the right adjusts the length of blade spring. The right figure is an experimental result. The resolution of the measurement system
is 0.2 (g) and the maximum relative error is less than 0.5 % of the sample masses.
|
>
