IROS 2018 Papers

Recently, as the rise of deep reinforcement learning, it not only can help the robot to convert the complicated environment scene to motor control command directly but also can accomplish the navigation task properly. In this paper, we propose a novel structure, where the objective is to achieve navigation in large-scale indoor complex environment without pre-constructed map. Generally, it requires good understanding of such indoor environment to make complex spatial perception possible, especially when the indoor space consists of many walls and doors which might block the view of robot leading to complex navigation path. By the proposed distributed deep reinforcement learning in different local regions, our method can achieve indoor visual navigation in the aforementioned large-scale environment without extra map information and human instruction. In the experiments, we validate our proposed method by conducting highly promising navigation tasks both in simulation and real environments.

Sliding window approaches have been widely used for object recognition tasks in recent years [19], [4], [5], [18]. They guarantee an investigation of the entire input image for the object to be detected and allow a localization of that object. Despite the current trend towards deep neural networks, sliding window methods are still used in combination with convolutional neural networks [22]. The risk of overlooking an object is clearly reduced compared to alternative detection approaches which detect objects based on shape, edges or color. Nevertheless, the sliding window technique strongly increases the computational effort as the classifier has to verify a large number of object candidates. This paper proposes a sliding window approach which also uses depth information from a stereo camera. This leads to a greatly decreased number of object candidates without significantly reducing the detection accuracy. A theoretical investigation of the conventional sliding window approach is presented first. Other publications to date only mentioned rough estimations of the computational cost. A mathematical derivation clarifies the number of object candidates with respect to parameters such as image and object size. Subsequently, the proposed disparity sliding window approach is presented in detail. The approach is evaluated on pedestrian detection with annotations and images from the KITTI [10] object detection benchmark. Furthermore, a comparison with two state-of-the-art methods is made. Code is available in C++ and Python https://github.com/julimueller/disparity-sliding-window.

Physical connectors with self-aligning geometry aid in the docking process for many robotic and automatic control systems such as robotic self-reconfiguration and air-to-air refueling. This self-aligning geometry provides a wider range of acceptable error tolerance in relative pose between the two rigid objects, increasing successful docking chances. We present a new method for computing the error range (or area of acceptance) for a pair of rigid connector objects with self-aligning geometry capable of higher dimensional analysis which was previously limited to three. The method is based on the configuration space obstacle model, which gives us a representation of the space of contact states between the two objects. Using an approach direction as analogous to gravity, and assuming the target docked configuration is stable, the set of misaligned points that lead to docking is the target configuration's watershed for an arbitrarily dimensioned configuration space obstacle. It is well known that the watershed of a height map on a discrete grid can be found using any number of algorithms from image segmentation. We present an implementation based on Meyer's flooding algorithm to determine this watershed and measure the AA for simple connectors in 2D and 3D. Results are presented for systems including unconstrained motion in SE(2) and motion constrained to four dimensions (ie. x,y,z,pitch) in SE(3).

Robots often have to deal with the challenges of operating in dynamic and sometimes unpredictable environments. Although an occupancy map of the environment is sufficient for navigation of a mobile robot or manipulation tasks with a robotic arm in static environments, robots operating in dynamic environments demand richer information to improve robustness, efficiency, and safety. For instance, in path planning, it is important to know the direction of motion of dynamic objects at various locations of the environment for safer navigation or human-robot interaction. In this paper, we introduce directional statistics into robotic mapping to model circular data. Primarily, in collateral to occupancy grid maps, we propose directional grid maps to represent the location-wide long-term angular motion of the environment. Being highly representative, this defines a probability measure-field over the longitude-latitude space rather than a scalar-field or a vector-field. Withal, we further demonstrate how the same theory can be used to model angular variations in the spatial domain, temporal domain, and spatiotemporal domain. We carried out a series of experiments to validate the proposed models using a variety of robots having different sensors such as RGB cameras and LiDARs on simulated and real-world settings in both indoor and outdoor environments.

We propose the Dexterous Manipulation Graph as a tool to address in-hand manipulation and reposition an object inside a robot's end-effector. This graph is used to plan a sequence of manipulation primitives so to bring the object to the desired end pose. This sequence of primitives is translated into motions of the robot to move the object held by the end-effector. We use a dual arm robot with parallel grippers to test our method on a real system and show successful planning and execution of in-hand manipulation.

There has been an increasing interest in mobile manipulators that are capable of performing physical work in living spaces worldwide, corresponding to an aging population with declining birth rates with the expectation of improving quality of life (QoL). Research and development is a must in intelligent sensing and software which will enable advanced recognition, judgment, and motion to realize household work by robots. In order to accelerate this research, we have developed a compact and safe research platform, Human Support Robot (HSR), which can be operated in an actual home environment. We assume that overall R&D will accelerate by using a common robot platform among many researchers since that enables them to share their research results. Currently, the number of HSR users is expanding to 33 sites in 8 countries worldwide (as of February 15, 2018). Software and technical knowledge of all users is shared through a community website. HSR has been adopted as a standard platform for international robot competitions such as RoboCup@Home and World Robot Summit (WRS). HSR is provided to participants of those competitions through public offering. In this paper, we describe HSR's development background, and technical detail of its hardware and software. Specifically, we describe its omnidirectional mobile base using the dual-wheel caster-drive mechanism, which is the basis of HSR's operational movement and a novel whole body motion control system. Finally, we describe the results of utilization in RoboCup@Home and field tests in order to demonstrate the effect of introducing the platform.

A pneumatically-driven flexible finger equipped with a flexible sensor is realized for improving the performance of the soft robotic hand. First, we propose a flexible angle estimation sensor. This sensor measures the change in the amount of light passing through polyurethane material and estimates the angle with high repeatability. Next, we design a flexible finger that makes this sensor easy to incorporate. The flexible fingers are produced with a multi-material 3D printer that can use flexible material. The flexible finger can accommodate the proposed flexible sensor within it. It is possible to place the sensor's signal line in the air pressure pipeline. Because the flexible finger is produced with a 3D printer, variations in each model's characteristics are small as compared with manufacturing through molding. In this paper, we show an improvement of positional accuracy in the proposed flexible finger using angle feedback control from the proposed sensor. The effectiveness of this sensor is also shown to solve the problem of vibration problems for the flexible finger during high speed motion.

The dolphin kick is a swimming style characterized by undulation of the body. As a platform for swimming research, we have developed a musculoskeletal humanoid robot called Triton. Triton has a flexible spine with erector spinae muscles and a stiffness adjustment system for lumbar joints. The musculoskeletal body includes biarticular and polyarticular muscles, providing multi-joint coordination. The robot is actuated by pneumatic muscles, yielding lightweight and inherently waterproof properties. The compliance of the joints allows interactions between body and fluid similar to those of human swimming. This study presents the design concept of Triton and experimental results from a water tank test. We compare the results with simulation and human movements reported in literature. The results show that the musculoskeletal swimming robot has similar cycle trends in joint angle and thrust force.

For decades, various robotic grippers have been developed due to its necessity for the robotic manipulators. In case of the conventional robotic grippers with rigid components, an underactuated mechanism was required to satisfy gripping motion. Recently, soft grippers have been studied actively, which have realize bending motion with a simple morphological structure itself and inherent compliance to the environment. In this field of study, it has been rarely investigated to improve the fingertip force and actuation speed with specified design parameters. Thus, in this study, a hybrid gripper, which consists of both soft and rigid components, was suggested based on the key design principles: 1) the ratio of rigid parts against the soft chamber, 2) the cross-sectional shape of the chamber. The suggested principles were verified using the finite element methods (FEMs). As a result, the improved performance of the hybrid gripper was verified in terms of the fingertip force and the actuation speed, compared with the performance of the previously developed soft pneumatic actuators (SPAs). As an application, the three-fingered gripper was manufactured and tested by grasping different types of objects.

This paper presents a high-quality wide-angle fovea lens, i.e., the WAF lens, for the autonomous robot's and vehicle's super-sensing vision system. The WAF lens is well-known in the field of robotic vision with respect to its unique design concept, biologically-inspired from a visual system of the primates. The WAF lens achieves the following two conflicting properties in imaging simultaneously: (1) wide field of view (FOV) and (2) high magnification factor (although only the central FOV achieves it partially). In this paper, the authors designs the WAF lens for the high-resolution photosensitive imaging chip more than 3M pixels. For this design, we decide the following targets on the assumption of applying this WAF lens for the stereo vision system: (1) The WAF lens can measure a very far distance over 100m ahead from the imager accurately. (2) The WAF lens can observe approximately 100-degree wide FOV on the same time. We produce a prototype of this WAF lens with much higher optical performance than our previous developments. The compound system of the prototype includes four aspherical surfaces in its front part to project enough bright images so that the WAF lens is available not only at daytime but also in dark situations at night. The authors experiment and demonstrate the projection tests using the prototype, and discuss about the results as the inspection of this challenging development.

This paper presents a lightweight (<;250 g) and low-cost (<;350 USD) biomimetic prosthetic hand with two actuators embedded in the palm. One of them is employed for flexion/extension of the five digits, and the other one is used for the adduction/abduction of thumb. Thus, the hand can achieve major grasping tasks that account for about 85% of activities in daily life. The unique transmission provides various advantages such as a compact structure, weight saving, and short driven distance. Furthermore, by using 3D printing technology, most parts of the prosthetic hand were made to be much lighter and have a humanlike appearance, compared with conventionally manufactured artificial hand. Finally, the performance and practical applicability of the proposed design was verified experimentally through both of a motion verification and an intuitive control test by a healthy subject and a transradial amputee.

In this paper, a novel mobile robot developed to perform Curling sports is introduced. The developed robot is a Stone Throwing Robot (STR) for Curling that can travel on the ice with wheels and throw a stone as well as make curls of the stone. The STR is developed as a robot component of an Artificial Intelligence(AI) system that can autonomously play the curling sport. The proposed STR can throw a stone at any desired speed and in any desired direction, which are determined by the AI system. To achieve this precise driving of the STR and throwing of the stone, two dimensional drive control is developed for the STR, which consists of 1) anti-slip control for high traction, 2) precise velocity control and 3) high accuracy heading angle control. In addition to the conventional PID controller, model-based feedforward control, Model Following Control (MFC) for the anti-slip control of the wheel on the ice and Yaw Moment Observer (YMO) for the robust heading angle control are applied as key technologies for the STR driving. The design configurations of the STR to achieve the detection of its own location and throwing/curling of the stone is proposed in this paper as well as the detail of the precise driving control.

This paper presents a novel method for generating level-ground walking for a rimless wheel with a controlled wobbling mass. Our rimless wheel achieves level-ground walking by simply controlling the wobbling mass attached to the wheel. We mathematically demonstrate that the controlled wobbling mass generates propulsive effects for the rimless wheel. The walking speed of the rimless wheel can be changed by varying the amplitude of the wobbling mass: thus slow walking to high-speed walking can be realized for the wheel. Moreover, we have developed a robot based on a rimless wheel to show effectiveness of our proposed methods. We then analyze the walking properties with respect to the physical parameters and control parameters of our robot through numerical simulation.

Motor skill learning is fundamental in many physical activities of human. In the processes of learning of motor skills, learners often receive visual or physical information about postures from teachers. However, the information about postures usually cannot be transmitted precisely. In this paper, we propose a motion teaching system to transmit teachers' motion to learners directly by using a motion capture and an assist suit. The assist suit, which has a pneumatic artificial rubber muscle (PARM) as an actuator, was designed to move a learner's hip joint with less loss of assistive force and less constraint of motion. Hip joint motion of a teacher can be transmitted to the assist suit by master-slave control. In addition, to compensate the delay of the PARM, posture of the teacher is predicted before the occurence by a recurrent neural network by using electromyogram signals and the past joint angle. We confirmed the system can transmit a teacher's motion to a learner in real time, and with the neural network, the delay of the learner's motion could be suppressed to approximately 0.1s, which is enough to feel visual and physical information synchronous. Therefore, the proposed motion teaching system would have the ability to transmit teachers' motion to learners visually and physically with precision sufficient to facilitate skill transmission.

This paper presents an improved design approach for the construction of a Magneto-Rheological (MR) clutch intended to be used in a prospective 5 degrees of freedom robot. The MR clutch features embedded Hall sensors for intrinsic torque control. After a brief description of the MR clutch principles, the details of the mechanical design are discussed. Simulation and preliminary experimental results demonstrate the main characteristics and advantages of the proposed MR clutch.

In this paper, a dual-arm robot system for high-speed manipulations, which is named LIMS2-AMBIDEX and is developed to compete in the IROS2018 Robotic Challenge, is presented. It has two seven-degrees-of-freedom (DO FF) lightweight arms, a three-DOF head, and a one-DOF gripper to manipulate foldable objects. Because all the heavy actuators are placed at the shoulder, it has remarkably low mass beyond the shoulder (2.63 kg), which guarantees an inherent safety at high speeds. Utilizing tension-amplification mechanisms, the high stiffness and strength are achieved, and thus it has the control performance comparable to conventional industrial manipulators. A unique three-DOF wrist mechanism, whose motions directly represent the quaternion values of the joint orientation, can manipulate objects without singular points in the entire range of motion. In order to utilize the object's inertia during rapid manipulation, the gripper was specially designed: it has a one-DOF finger to grasp the upper rib of the foldable fan and two supporting forks to grasp the bottom rib stably. For real-time performance and increased scalability, a software framework was developed based on Robot Operating System (ROS). The real-time capability is achieved by using the real-time development framework Xenomai and the high-speed communication protocol EtherCAT. As most of the algorithms are implemented in the distributed nodes using ROS, it is convenient to expand, improve, and replace the algorithms. Consequentially, the entire motion of the Fan Robotic Challenge Phase I Modality B required 1.05 s, which is substantially faster than a similar manipulation by most humans.

Currently, serious accidents are caused frequently by the aging of sewer pipes. Therefore, to inspect sewer pipes, we developed a peristaltic crawling robot that reproduces the locomotion of an earthworm. This robot can drive for more than 100m, and it can be used for the maintenance of sewer pipes (100A pipes). However, the speed of the robot is low. There are two causes. First, the units of the previous robot have steps of artificial muscle fastening. These steps increase the diameter of the artificial muscles. A smaller diameter of the artificial muscles is advantageous for the speed of the robot inside the pipes. Second, the previous robot has slow air response. In this study, we used large-sized solenoid valves to overcome this drawback.

This study presents principle of the camber-flat wing structure conversion mechanism, which is inspired by a dragonfly, and its applicability to MAV. The camber-flat wing structure convert mechanism makes MAV flight using asymmetric flapping pattern through control of angle of attack without complicate structure. This mechanism was inspired from the dragonfly's feature that the camber structure of the wing increases the rigidity of wing structure and makes dragonfly has asymmetric flapping pattern. Experimental results show that MAV has asymmetric flapping pattern that can more stable flight performance when hovering flight with a camber structure and superior performance when the forward flight with a flat structure. The average lift force in the camber wing structure was 0.02N, the average thrust force was 0.02N and the average lift force was 0.011N in the flat wing structure at 20 Hz flapping frequency.

In clinical practice, magnetic resonance imaging (MRI) is used to locate a lesion/tumor for bone biopsy in children. However, there is a lack of MR-compatible tools that can be used simultaneously during imaging and biopsy while maintaining surgical accuracy and safety. The Pediatric Surgery Robot (PSR) platform is a 5-DOF robot with a modular tool interface. For the case of bone biopsy, a Bone Biopsy Tooling (BBT) is attached. It is designed to fit within a Philips Achieva 3.0T MRI bore and carry a modified titanium bone biopsy needle. A surgical pre-planning and control interface has been developed for joint and Cartesian level control. The PSR-BBT has demonstrated 1.65 +/- 1.77 mm accuracy in Cartesian control in free space. The PSR-BBT can generate 12.46 +/- 0.32 N of axial force while drilling at a speed of 30 rpm, which is sufficient for cortical and cancellous bone phantoms. Under MRI testing (T1-FFE, T1-SE, T2-FFE and T2-TSE scans), the system demonstrated less than 33% signal-to-noise ratio variation while drilling and a 0.46% geometric distortion while powered on without significantly impacting MRI guidance in situ. These results show that the PSR-BBT can allow the user to simultaneously image and perform the biopsy and presents the PSR as a viable platform for MR-guided robotic surgery.