IROS 2018 Papers

This video shows the main features of Hammer, a tablet-based end-user interface for industrial robot programming, in a real environment: a robotic cell created for the Hephestos European project. Hammer is an Android application that makes easier to program tasks for industrial robots like polishing, milling or grinding. It is based on the Scratch programming language, but specifically design and created for Android OS. It is a visual programming concept that allows non-skilled operators to create programs. The application allows to monitor the tasks while it is being executed by overlapping real time information through augmented reality. The application includes a teach pendant screen that can be customized according to the operator needs at every moment. The application is designed for online programming and reprogramming; easy use of learn-by-demonstration methods; easy connection with the robot control and sensors systems; and safety-system integration. It aims to be intuitive, easy to use, and simple. The application has four main parts: customized teach pendant, robot programming IDE and simulator, manual-guidance interface and augmented-reality-based-monitoring system.

In recent years, we have observed a clear trend in the rapid rise of autonomous vehicles and robotics. One of the core technologies enabling these applications, Simultaneous Localization And Mapping (SLAM), imposes two main challenges: first, these workloads are computationally intensive and they often have real-time requirements; second, these workloads run on battery-powered mobile devices with limited energy budget. Hence, performance should be improved while simultaneously reducing energy consumption, two rather contradicting goals by conventional wisdom. Previous attempts to optimize SLAM performance and energy efficiency usually involve optimizing one function and fail to approach the problem systematically. In this paper, we first study the characteristics of visual inertial SLAM workloads on existing heterogeneous SoCs. Then based on the initial findings, we propose π-SoC, a heterogeneous SoC design that systematically optimize the IO interface, the memory hierarchy, as well as the the hardware accelerator. We implemented this system on a Xilinx Zynq UltraScale MPSoC and was able to deliver over 60 FPS performance with average power less than 5 W.

Limited perception capabilities underwater shrink the envelope of effective localization techniques that can be applied in this environment. Long-term localization in six degrees of freedom can only be achieved by combining different sources of information. A multiple vehicle underwater localization solution, for localizing an underwater mining vehicle and its support vessel, is presented in this paper. The surface vessel carries a short baseline network, that interact with the inverted ultra-short baseline, carried by the underwater mining vehicle. A multiple antenna GNSS system provides data for localizing the surface vessel and to georeference the short baseline array. Localization of the mining vehicle results from a data fusion approach, that combines multiple sources of sensor information using the Extended Kalman Filter (EKF) framework. The developed solutions were applied in the context of the ¡VAMOS! European project. Long-term real time position errors below 0.2 meters, for the underwater machine, and 0.02 meters, for the surface vessel, were accomplished in the field. All presented results are based on data acquired in a real scenario.

Preprogramming of tasks still plays an important role in complex robotic systems despite the advances in automated planning and symbolic learning. Often, it is desired that end-users implement further tasks to adapt the robotic application to their needs. These user-defined tasks have to meet safety and integrity constraints for protecting the robotic platform and its users. We introduce a verifiable task specification language (vTSL) that enables to automatically prove that a task specification satisfies a set of predefined or task-specific constraints. We illustrate our approach using an example of a self-driving vehicle for intra-logistics and report experiences with two commercial applications.

Limitations with the on-board computational resources installed on untethered robots such as humanoids and mobile robots in general affects significantly the performance and capabilities of these machines. An approach to address this issue is to make use of the cloud robotics concept and take advantage of the extensive computational resources of the cloud. XBotCloud is a recently developed component of the XBot framework. It tackles the above challenges by introducing the tools and mechanisms to enable users and robots to exploit the computational resources of the cloud allowing the execution of services with low, soft or hard Real-Time execution/communication performance. The latter is ensured thanks to the functionality provided by the XBotCore Real-Time cross-robot software component of the XBot framework. XBotCloud addresses also one of the main challenges related with cloud robotics: security. To avoid remote attacks it takes advantage of the Amazon Web Services (AWS)Cloud Security and it uses an internal VPN Network to handle the connectivity between the robot and the cloud server. The full implementation of the framework is presented and its functionality is demonstrated in realistic tasks involving pipelines that mix the execution of cloud services with moderate execution time constraints and Real-Time modules running on the robot local control unit. XBotCloud performances and cross-robot flexibility are experimentally validated on two different robotic platforms, the WALK-MAN humanoid and the CENTAURO upper body/full-body.

This work provides a framework for a workspace aware online grasp planner. This framework greatly improves the performance of standard online grasp planning algorithms by incorporating a notion of reachability into the online grasp planning process. Offline, a database of hundreds of thousands of unique end-effector poses were queried for feasibility. At runtime, our grasp planner uses this database to bias the hand towards reachable end-effector configurations. The bias keeps the grasp planner in accessible regions of the planning scene so that the resulting grasps are tailored to the situation at hand. This results in a higher percentage of reachable grasps, a higher percentage of successful grasp executions, and a reduced planning time. We also present experimental results using simulated and real environments.

Provides an abstract for each of the workshop presentations and may include a brief professional biography of each presenter. The complete presentations were not made available for publication as part of the conference proceedings.

This paper presents a novel wireframe map structure for resource-constrained robots operating in a rectilinear 2D environment. The wireframe representation compactly represents geometry, in addition to transient situations such as occlusions and boundaries of unexplored regions. We formulate a particle filter to suit this sparse wireframe map structure. Functions for calculating the likelihood of scans, merging wireframes, and resampling are developed to accommodate this map representation. The wireframe structure with the particle filter allows for severe discrete map errors to be corrected, leading to accurate maps with small storage requirements. We show in a simulation study that the algorithm attains a map of an environment with 1 % error, compared to an occupancy grid map obtained with GMapping which attained 23% error with the same storage requirements. A simulation mapping a large environment demonstrates the algorithms scalability.

In the received signal strength (RSS) based localization, a model-based calibration approach has been usually done by relating RSS-to-distance among anchor nodes. In this paper, an improved calibration method is proposed. For that purpose, RSS and estimated distance between any pairs of an-chor/unknown nodes is considered under the linear regression model. Unfortunately in this model, partial input elements are erroneous due to the inaccurate localization of unknown nodes. To obtain its solution under consideration of such an error, the weighted total least squares (WTLS) techniques are employed here. With the help of the WTLS techniques, several errors involved in the model can be effectively compensated. To show the efficiency of the proposed calibration, it is combined with several localization algorithms and its performance is verified by various simulations. The results show that the proposed calibration can give a very similar localization performance to that of each localization algorithm when true model parameters are known.

In a human-robot interaction (HRI) device that performs physical collaboration operations in constant contact with the user, admittance control and impedance control are generally used. Since the two controllers exhibit opposite performances depending on the stiffness condition, controllers capable of dealing with various magnitudes of stiffness are required. As such, this study proposes hybrid control that adjusts the control distribution ratios of admittance control and impedance control based on the operating frequency analysis to react to the user intention and various stiffness conditions in real time. The proposed controller algorithm exhibited lower overshoot than impedance control in the step input response simulation, faster response speed compared to admittance control in the response simulation for 0-5 Hz input frequencies, and the smallest vibration magnitude and number of vibrations in the case of a virtual wall collision, resulting in improved performance compared to existing control methods.

This study reports the initial testing of a gait exoskeleton for Spinal Muscular Atrophy (SMA) patients having variable muscle strength with no balance or ambulation capabilities. To improve the quality of life of such patients, a pediatric gait exoskeleton was developed. The ATLAS exoskeleton has 8 active degrees of freedom (DOF): 2 at the hip (adduction/abduction and flexion/extension), 1 at the knee and ankle joint for flexion and extension. A feasibility test was performed to gauge the initial response of the patients. This study demonstrates that the exoskeleton was able to provide gait assistance and sit-to-stand movements effectively to the subjects. This kind of wearable exoskeleton will play a key role in the rehabilitation of SMA patients and delay further metabolic degeneration in the future.

In this paper, we propose methods for walking on a steep slope using a rope by a humanoid robot. There are two difficulties for walking on a steep slope without a rope. First, range of motion of ankle joints get limited. Second, feet of a robot slip on a steep slope. For these problems, using a rope is effective solution because the robot can receive enough friction force from the slope and walk on a steep slope by pulling a rope with proper tension. In addition, the robot pulling a rope on a slope can relax limitations of ankle joints. Therefore, we propose methods to determine tension of a grasped rope by solving a linear least-square problem considering deformability of a rope. With these methods, a life-size humanoid robot HRP-2 could walk on a steep slope which angle is 40 degree.

The aim of this study was to develop a walking assistance device by controlling trunk and pelvis motion. Human gait motion is a whole-body exercise that includes both lower body and upper body motion. Therefore, focusing on trunk and pelvic rotational motion, assist and resist the walking by applying external forces. By introducing periodic input control, control the timing of the supplementary force, and assist the walking. As an experimental result, the walking assistance improves gait by decreasing energy loss, and the walking resistance applies a load during walking.

Dynamic balance control for humanoid robots encounters difficulties such as stability, speed, and smoothness. In most of the previous studies, joints act as controller of the Center of Mass (CoM)supported using a simplified mathematical model. Then, the stability of the motion is guaranteed using the Zero Moment Point (ZMP)stability criterion. In this video, a humanoid robot [1] will carry a tray secured to the wrist and the objects to be transported will be placed on the tray. This condition implies that the object is not grasped and therefore, the robot arm will be the only point of support of the object through the tray. Thus, the manipulation control system must be able to detect the stability of the object and act according to the different perturbations applied to it. A 3D balance control system for non-grasping tasks is presented and it is based on the ZMP criterion and 3D inverted pendulum equations. The perception system required is based on the use of Force-Torque sensors [2], computer vision [3], and their integration. The effectiveness of the proposed approach is being investigated with the humanoid robot TEO.

Many industrial facilities require periodic visual inspections. Often the points of interest are out of reach or in potentially hazardous environment. Multi-copters are ideal platforms to automate this expensive and tedious task. This video presents a system that enables a human operator to teach a visual inspection task to an autonomous aerial vehicle by simply demonstrating the task using a tablet. The system employs the Google Tango visual-inertial mapping framework as the only source of pose estimates, thus enabling operation in GPS-denied environments. In a first step the operator records the desired inspection path using the tablet. Inspection points are automatically inserted if the operator pauses, holding a viewpoint. The mapping framework then computes a feature-based localization map, which is shared with the robot. After take-off, the robot estimates its pose based on this map and plans a smooth trajectory through the way points defined by the operator. Furthermore, the system is able to track the global pose of other robots or the operator, localized in the same map, and follow them in real-time, while avoiding collision. This was demonstrated in the second part of the video, where the robot is following the operator in real-time through a hedge maze.

We present a location-specific method to visually track the positions of observed vehicles based on large-scale crowd-sourced maps. We equipped a large fleet of cars that drive around cities with camera phones mounted on the dashboard, and performed city-scale structure-from-motion to accurately reconstruct the trajectories taken by the vehicles. We show that these data can be used to first create a system enabling high-accuracy localisation, and then to accurately predict the future motion of newly observed cars in the camera view. As a basis for the method we use a recently proposed system [1] for unsupervised motion prediction and extend it to a real-time visual tracking pipeline which can track vehicles through noise and extended occlusions using only a monocular camera. The system is tested using two large-scale datasets of San Francisco and New York City containing millions of frames. We demonstrate the performance of the system in a variety of traffic, time, and weather conditions. The presented system requires no manual annotation or knowledge of road infrastructure. To our knowledge, this is the first time a perception system based on a large-scale crowd-sourced maps has been evaluated at this scale.

This paper describes a vision-based target tracking method for a skid-steer vehicle. With the development of deep reinforcement learning, many researchers have tried to generate an end-to-end policy to control the mobile robot from a raw pixel image data. However, the action in most research only concerns high-level decisions such as go straight, turn left and right. High-level decisions alone are not sufficient to precisely control platforms such as a skid-steer vehicle due to the lack of steering mechanism. Thus, unlike existing work, we aim to control the motor command for the wheels directly. To this end, we employ guided policy search (GPS) based on the general kinematic slip model for the skid-type robot. Furthermore, to prohibit the target from getting out of the camera field of view (FOV) in the training phase, we update local policy optimization with a FOV constraint and perform a pre-training to make the initial policy more efficient. Our method allows the skid-type robot to automatically acquire the vision-based tracking policy while local policies satisfy the FOV constraint during the training phase. We evaluate our method through both simulation and experiment with a skid-steer mobile robot. Finally, we test the performance of learned policy with a moving target in a new environment.

This paper examines techniques for real-time terrain classification and solar irradiance mapping for outdoor, solar-powered mobile robots using a vision-based Artificial Neural Network (ANN). This process is completed sequentially. First, terrain classification is completed by extracting key features from visual-spectrum images captured from an on-board camera using Haar wavelet transform to identify both color and textural information. These features are then classified using an ANN to identify grass, concrete, asphalt, gravel, and mulch. Using the terrain classes, the image is then analyzed using concepts from high dynamic range imagery to establish the solar irradiance map of the area. In this way, our sequential methodology presented allows unmanned vehicles to classify the terrain and map the irradiance of a given area with no prior knowledge. Whereas, the terrain classification can be used in determining energy consumption or traversability criteria and the irradiance map can be used to estimate the energy harvesting capabilities.

This paper presents an approach to surgical tool tracking using stereo vision for the da Vinci® Surgical Robotic System. The proposed method is based on robot kinematics, computer vision techniques and Bayesian state estimation. The proposed method employs a silhouette rendering algorithm to create virtual images of the surgical tool by generating the silhouette of the defined tool geometry under the da Vinci® robot endoscopes. The virtual rendering method provides the tool representation in image form, which makes it possible to measure the distance between the rendered tool and real tool from endoscopic stereo image streams. Particle Filter algorithm employing the virtual rendering method is then used for surgical tool tracking. The tracking performance is evaluated on an actual da Vinci® surgical robotic system and a ROS/Gazebo-based simulation of the da Vinci® system.