The Dynamic Window Approach (DWA) method, due to its constant acceleration model with velocity invariance assumption, the generated paths concentrate the arrival points in a narrow range, and it is impossible to control jerk. To solve these problems, we propose a method for generating variable velocity path based on jerk models. First, we changed DWA's constant acceleration model to constant jerk model, which allows us to limit the maximum jerk as a parameter. Then, we use constant jerk model to generate variable velocity sequences to obtain a wider range of paths. Finally, we added the jerk value as a term to the evaluation function to prevent consistently large jerk. We compared the proposed method with the original method in three different simulator environments and in the real world, verifying the effectiveness of the proposed method.
Due to their safe interaction and dexterous movements, continuum robots have potential applications in care, welfare, and search and rescue. However, they often struggle with their limited reach and payload. To address this problem, we focus on woodpeckers, which have an excellent elongated manipulator, or tongue, capable of extending and bending without sagging due to gravity. We propose a novel rod-driven continuum that combines a woodpecker-inspired extension mechanism and convergent rod-path routing. The proposed robot successfully changed the effective length from 0 mm to 600 mm and improved the stiffness up to 12 times.
Because the demand for ultrasound examinations has increased recently, training opportunities also need to be increased. Therefore, we developed a training simulator for beginners that can be used at home to learn abdominal anatomy, without using ultrasound machines. A probe-shape controller and training display that has a cross-sectional image corresponding to the position and posture of the controller were developed, and the effectiveness of the simulator was confirmed through performance evaluation and evaluation by physicians.
In this paper, we propose a method to reduce water consumption by using continuous transfer in a peristaltic transfer system for transporting feces under microgravity conditions such as in space or on a lunar base. Recent long-term manned space exploration requires a ``Environmental control and Life Support System'' due to the difficulty of resupply. The authors propose a method to reduce wa-ter consumption by using a peristaltic transfer system for continuous transfer. The method was test-ed in a simulated feces transfer experiment to evaluate its effectiveness in reducing water consump-tion and transfer performance compared to conventional methods.
For robots to behave intelligently in the real world, they need to acquire knowledge through interactions with the environment. In this study, we investigated the effect of grasp reflex behaviors, similar to those in human infants, on a model that learns complementary actions and object concepts based on the robot's own curiosity. Simulation experiments showed that early-stage grasp reflexes have a positive effect on sensory information gathering and object concept learning. However, robots without curiosity tend to forget their actions with the disappearance of reflexes, leading to the collapse of their internal representations of object concepts. These results suggest that both curiosity and reflexive behavior are important for actions and concept learning.
Learning from Demonstration, in which robots are taught based on human demonstrations, requires techniques for partitioning complex tasks into simple skills. In assembly tasks, contact state is an important feature that describes skills, but it is affected by variations in the measurement coordinate system and trials. In this paper, we propose a contact state estimation method based on principal component analysis and force coordinate transformation. Features of assembly operations are extracted by principal component analysis of position information and coordinate transformation of force information. By using the extracted features, environment-independent contact state estimation is possible.
We have evaluated an effectiveness of information presentation from robots to human using mixed reality, supposing situations in which humans and robots live and cooperate with each other. We investigated the relationship between presented-contents of the robot's motion target, i.e, i. position, ii. direction, and iii. combination of position and direction, and reaction time. We confirmed that content iii. is superior in reaction time (0.90±0.22[s]). Also, according to preliminary investigation of Mental Workload, no difference was found due to the increase in the amount of information from content i. to content iii.
This paper describes a method to create a dataset for object detection from aerial drone images using UnrealEngine5. Creating a dataset from aerial drone images in real-world is challenging due to constraints such as legal and safety regulations. A virtual environment enables the creation of datasets in any desired situation. It is also possible to simulate the aerial perspective of drones for virtual dataset generation. The experimental results for detection accuracy of models produced by the virtual dataset and by the existing real-world dataset show that the virtual dataset is capable of detecting objects in real-world.