RESEARCH TOPICS
-Autonomous Flight
-Aerial Manipulation
-Autonomous Robot Learning
-Visual Navigation & Active Perception
-Multi-robot Systems
-New Aerial Platforms
Autonomous Flight
For the autonomy of aerial vehicles in various applications, we develop planning strategies which consider both the flight efficiency and safety. We are also interested in applying those strategies for complex aerial systems equipped with robotic arms or slung loads. Also, motion that promotes an active perception for vision-equipped drones is designed for the tasks such as visual servoing, videography and navigation. Adding to motion planning, we apply robust and adaptive control methods for the stable operation of drones with analysis of reachability or convergence.
Youtube playlistAerial Manipulation
Aerial manipulators refer to the composition of UAVs and manipulators such as suspended cables or robotic arms with multi degrees of freedom. They aim to actively interact with the surrounding structures and perform sophisticated tasks that cannot be accomplished by bare UAVs, such as aerial grasping of an object and operating a spatial structure. They pose exciting yet challenging problems due to a potential loss of stability caused by the manipulator movement, physical adjacency to surroundings, and limited endurance. For successful aerial manipulation, we develop a dedicated controller and an optimal trajectory planner considering obstacles in the workspace. We further extend our approaches to cooperative scenarios to perform missions impossible for a single aerial manipulator.
Youtube playlistAutonomous Robot Learning
We are interested in capabilities to overcome various real-world problems not by human engineering but by robot learning. We study the various types of autonomous robot learning ranging from understanding visual sensory input to self-developing behavior. These topics include visual recognition, reinforcement learning, imitation learning, transfer learning, and their variants. Recent success of deep learning provides most interesting tools, but more studies are needed especially because we always contemplate applications to actual robots that have physical constraints and limited resources. We hope to contribute toward robots that move out of a laboratory and enhance our life.
Youtube playlistVisual Navigation & Active Perception
Knowledge of own position and pose is essential for unmanned vehicles to fulfill their missions autonomously. We are interested in obtaining an accurate navigation solution using perceptive sensors. Specifically, we focus on utilization of monocular, stereo, and RGB-D cameras for estimating the 6-dof pose (i.e., orientation and translation), and improving the performance of the algorithm in the presence of dynamic elements such as illumination changes or moving objects. We also conduct research on a new type of vision sensor called an event camera, which is motivated by the human eye and generates events asynchronously when each pixel detects log-intensity changes of light. We are developing a new navigation algorithm for harsh environments, which can utilize its advantages such as low latency, high time resolution, and high dynamic range.
Youtube playlistMulti-robot Systems
Deploying a multi-robot system to accomplish a common goal offers several advantages, such as scalability, flexibility, and cost-efficiency of operation. For example, multi-agent informative planning can be utilized to construct local information maps such as climate, topographic maps, and city traffic. Despite their versatility, there remain many challenges including coordination between agents, communication limitations, and task allocation. In the coordination research, robots’ paths are optimized for operation in a real environment avoiding collisions. Considering communication limitations, distributed processing algorithms are investigated to achieve an efficient data exchange within limited network resources. We also study collaborative multi-robot navigation systems to improve overall localization performance.
New Aerial Platforms
We explore new types of aerial robot platforms that can overcome the inherent limitations of the conventional aerial robot platforms. We have developed a new type of multirotor platform that enables six controllable degrees-of-freedom flight with minimal structural heterogeneity compared to the conventional multirotor design. This new multirotor platform consists of upper and lower parts (or thruster and fuselage parts), and a unique mechanism that controls the relative attitude between the two parts. With this, the fuselage can control the translational and rotational motions independently of each other, allowing 6-degrees of freedom motion that was not possible with conventional multirotors. We are also investigating micro flapping-wing aerial robots with improved onboard control mechanism and learning-based controllers for stable autonomous flight.
Youtube playlist
