Series elastic tendon-drive robots have been proposed intended for safe physical human-robot interaction. While the tendon elasticity enables a safe task execution under hand contact conditions, it poses overshoot and vibration problems in tasks without hand contact. In this paper, we install an electromagnetic brake in the pulley at the middle of the cable path, in order to stabilize the vibration. This method enables a wide range of viscosity changes, because the elements are arranged so that the brake force is directly transmitted to the arm. Moreover, redesign of the joint structure is unnecessary. The elastic tendon with brake can be modeled as a parallel viscoelastic element. We propose a sliding mode controller by using the electromagnetic brake. High-speed friction control by the electromagnetic brake allows an oscillation-free positioning controller. Unlike conventional sliding mode controllers, softness (low stiffness) of the controlled arm is not lost because high-gain position feedback is not used. Reachability condition of the sliding surface is expressed in terms of the equilibrium and target positions of the arm angles. Control results are demonstrated experimentally.

We are currently researching an AI robot camera that has learned the shooting techniques used by broadcast cameramen, with the aim of applying it to golf broadcasts. Information such as the player's behavior measured by the environmental information measurement camera and the three-dimensional coordinates of the ball measured by the stereo camera is shared over the network, and multiple robot cameras work together to take pictures. Rather than the inorganic shooting caused by the feedback control of general robot cameras, it is possible to cooperatively shoot images that are as natural and realistic as if they were being taken by a human. In experiments, we confirmed that it was possible to accurately shoot players and caddies before a shot, track the ball during flying, and track players as they move. It was also confirmed that the trajectory could be drawn in real time using CG. The technology of this system was used for broadcasting at the Japan Golf Tour Championship.

In Japan, there are islands such as Nishinoshima that cannot be accessed due to volcanic activity. There is academic interest in how ecosystems recover after major eruptions. Because the topography and geology are unknown, and there are still dangers from fumarole and volcanic gas, a rover is needed to take the place of humans and land on land. Authors have developed a new rover that is lightweight enough to be transported by multicopter, capable of taking images of the site, removing obstacles, and collecting samples. In this paper, we report the configuration of the proposed vehicle, system configuration for remote control, mechanical design, and results of field experiments.

Joining soft and hard materials, such as silicone rubber and a plastic part, is an obstacle to inexpensive and simple prototyping. This study proposes a liquid infill (LIF) method for joining soft and hard materials. We selected the gyroid for the porous structure, which is well known as one of the infill structures of 3D printed parts made using fused deposition manufacturing. Our proposed manufacturing process only requires a 3D printer and silicone rubber molding equipment. The joining strength of samples made by LIF is evaluated with tension tests. The results indicate that material hardness affects tensile strength, while infill density had little effect. Meanwhile, the amount of strain at break decreased with increased infill density. In addition, we compared the tensile strength of samples made by LIF with samples joined by anchor and adhesive. As a result, the tensile strength of LIF exceeds the two other types of samples. Furthermore, we captured slice images using a CT scanner and confirmed that the silicone rubber was filled in gyroid structures. Finally, we fabricated the backbone mechanism and the finger mechanism.

In recent years, educational support robots have been attracting attention. In conventional robots, the leaner cannot learn in a self-directed learning environment where they decide the number of problems by themselves. As a remedy, we proposed the Problem-Number Negotiation Method by deciding the number of problems through a conversation with the learner before learning. However, this method has not been verified in terms of changes in the number of problems through long-term collaborative learning. In this paper, we examine the effect of long-term learning with a robot that uses proposed method on the number of problems by university students.

Using many agents with different characteristics is more effective than using a homogeneous agent to observe a vast environment persistently. In order for heterogeneous multi-agents to efficiently perform persistent observation of a changing environment, it is necessary to plan the most informative paths for each agent according to its observation capability. In this paper, we report the results of applying the path planning method for maximizing the amount of information obtained by Monte Carlo tree search to the mutual information map obtained from the estimated state of the environment and the observation ability of the agent.

In a previous study, the authors proposed a method to adapt the dynamic stability of walking, MoS, to walking with a cane. Specifically, the moment of equilibrium (CoP) point between the legs and the cane load was used as the boundary of the BoS. As a result, it was confirmed that the MoS tended to be higher at a specific cane tip position. In this study, a musculoskeletal model was developed for theoretical analysis. The musculoskeletal model consists of four links from the ground to the shoulder, upper arm, forearm, and cane; three joints (shoulder, elbow, and hand); and three muscles (deltoid, triceps, and extensor carpi ulnaris). Using this model, the cane load was estimated for the cane tip position. The results showed that the estimation accuracy was high in the range of 20 to 40 cm.