IROS 2019 Papers

Difficulties in capturing unique structures in the outdoor environment hinders the map-based Autonomous Vehicles (AV) localization performance. Accordingly, this necessitates the use of high resolution sensors to capture more information from the environment. However, this approach is costly and limits the mass deployment of AV. To overcome this drawback, in this paper, we propose a novel outdoor map-based localization method for Autonomous Vehicles in urban environments using sparse 3D lidar scan data. In the proposed method, a Point-to-Distribution (P2D) formulation of the Normal Distributions Transform (NDT) approach is applied in a Monte Carlo Localization (MCL) framework. The formulation improves the measurement model of localization by taking individual lidar point measurements into consideration. Additionally, to apply the localization to scalable outdoor environments, a flexible and efficient map structure is implemented. The experimental results indicate that the proposed approach significantly improves the localization and its robustness in outdoor AV environments, especially with limited sparse lidar data.

Among the various pneumatic actuators, thin McKibben muscles are particularly attractive owing to their high performance in terms of high contraction ratio, high flexibility, and productivity. One of the challenges for thin McKibben muscles and other pneumatic actuators is the fact that they generally need air hoses and air sources like compressors. This necessarily leads to bulky systems. To solve this problem, we propose and study a new approach to driving pneumatic actuators that exploits electrolysis/synthesis of water by a polymer electrolyte fuel cell (PEFC). This method could in-principle have high applicability to thin McKibben muscles. However, one challenge must still be addressed for the realization of an electrically driven thin McKibben muscle: the development of a tube-shaped soft PEFC. This paper proposes an electrically driven thin McKibben muscle with a tube-shaped soft PEFC realized through a flowing non-electrolytic plating method. First, the proposed thin McKibben muscle is briefly described. Then, a novel method for plating a tube-shaped soft PEFC using reflux is introduced, and its fundamental operation is tested. After the evaluation of the PEFC, we report a prototype realization of the proposed thin McKibben muscle. The developed prototype has a length of over 170 mm, a diameter of 4 mm, and high flexibility for bending. Finally, we elaborate on a driving experiment through which fundamental characteristics were experimentally evaluated. The prototype thin McKibben muscle succeeded in electrically controlling contraction and expansion motion with a flexible and bendable structure.

Soft contraction actuator has always been an important research field in soft robotics. This paper presents a novel soft contraction actuator which can realize high-ratio contraction when pressurized by pneumatic power. This actuator can be easily made by soft materials including rubber tube, one-way extensible cloth, inextensible cloth and wire. 2 types of structures are presented based on different working circumstances. When pressurized, the actuator can curve and then coil to a helical shape due to the different extensibilities of its opposite sides, a high contraction length and large contraction force can be achieved synchronously. Experiments are conducted to show the relationship between contraction force and contraction ratio of this actuator. The high contraction ratio of this actuator is revealed by comparing with McKibben actuator under the same condition, the maximal contraction ratio reaches 78% and the actuation stress is 8.36kPa.

Recent advances of actor-critic methods in deep reinforcement learning have enabled performing several continuous control problems. However, existing actor-critic algorithms require a large number of parameters to model policy and value functions where it can lead to overfitting issue and is difficult to tune hyperparameter. In this paper, we introduce a new off-policy actor-critic algorithm, which can reduce a significant number of parameters compared to existing actorcritic algorithms without any performance loss. The proposed method replaces the actor network with a set of action particles that employ few parameters. Then, the policy distribution is represented using state action value network with action particles. During the learning phase, to improve the performance of policy distribution, the location of action particles is updated to maximize state action values. To enhance the exploration and stable convergence, we add perturbation to action particles during training. In the experiment, we validate the proposed method in MuJoCo environments and empirically show that our method shows similar or better performance than the state-of-the-art actor-critic method with a smaller number of parameters. The experimental video can be found at http: //rllab.snu.ac.kr/multimedia.

Winged aerial robots represent an evolution of aerial manipulation robots, replacing the multirotor vehicles by fixed or flapping wing platforms. The development of this morphology is motivated in terms of efficiency, endurance and safety in some inspection operations where multirotor platforms may not be suitable. This paper presents a first prototype of compliant dual arm as preliminary step towards the realization of a winged aerial robot capable of perching and manipulating with the wings folded. The dual arm provides 6 DOF (degrees of freedom) for end effector positioning in a human-like kinematic configuration, with a reach of 25 cm (half-scale w.r.t. the human arm), and 0.2 kg weight. The prototype is built with micro metal gear motors, measuring the joint angles and the deflection with small potentiometers. The paper covers the design, electronics, modeling and control of the arms. Experimental results in test-bench validate the developed prototype and its functionalities, including joint position and torque control, bimanual grasping, the dynamic equilibrium with the tail, and the generation of 3D maps with laser sensors attached at the arms.

Flexible human-robot cooperation can be coordinated by dynamic task allocation, where team members observe task progress and repeatedly decide to execute outstanding operations from a commonly known task model. Such task-level coordination strategies do not cover cases, where a high level of abstraction for task modelling implies operations infeasible for one single agent due to limited capabilities. Task-level coordination can handle these operations by introducing additional operation-level coordination for more fine-grained interaction, e.g. to hand over a tool or provide a helping hand. We contribute an approach to classify operations regarding the interaction they require. Based on this classification, we show how robots can plan and issue communication to coordinate respective interaction within dynamic plan execution. Against the background of Smart Factories, a smartphone app is used as communication channel and evaluated in a user study.

Perception of the intention of humans prior to an interaction is a demanding skill during human-robot interaction (HRI). This skill is even more sought after during robot-initiated HRI. Initiating an interaction in an inappropriate situation can be avoided when robots are equipped with the ability to decide when to interact and when not to. Many of the existing systems investigate only a few characteristics of humans which demonstrate inner state of mind and are based on complex monitoring mechanisms which limit their use in most of the scenarios. This work presents an autoregressive model based on observable physical and emotional human cues to determine the level of interest displayed by a human towards an interaction with a robot. This model was implemented on a service robotic platform and the behavior of the robot was controlled using the model. The behavior of the robot was determined by means of proxemic approach and the nature of conversation with the human. The outcomes of the model were evaluated by analyzing user feedback in different situations inside a simulated social environment. Using the model, robot was given the ability to analyze the situation of its human user in an emotionally intelligent manner, prior to an interaction. The behavior of the model was reviewed by user feedback in order to validate the findings. Results of the experiment are presented and findings of the study are discussed.

For an autonomous mobile system such as an autonomous robot that moves throughout a city, semantic segmentation is important. Performing semantic segmentation under diverse conditions, in turn, requires 1) a robust ability to recognize objects in low-visibility environments, such as at night and 2) the ability to recognize objects that transmit visible light, such as glass and acrylic used in doors and windows. To satisfy these requirements, using RGB images and infrared images simultaneously is considered effective. Visibility and infrared transmission characteristics are different for different objects; therefore, merely entering them into the conventional semantic segmentation framework is not applicable. For example, when a pedestrian is present behind a glass, the visible image captures the pedestrian rather than the glass and the infrared image captures the glass. In this research, we propose a new semantic segmentation method having a three-stream structure, focusing on the difference in the transmission characteristics. This method extracts not only valid features for ordinary non-transparent objects but also features effective for the recognition of transparent objects by utilizing differences in objects to be imaged owing to transmission characteristics. Furthermore, we constructed a new dataset called “coaxials” for the visible and infrared coaxial dataset and demonstrated that we can obtain better segmentation performance compared with the conventional method.

We present a method for simultaneous localisation and wind turbine model fitting for a drone performing an automated surface inspection. We use a skeletal parameterisation of the turbine that can be easily integrated into a non-linear least squares optimiser, combined with a pose graph representation of the drone's 3-D trajectory, allowing us to optimise both sets of parameters simultaneously. Given images from an onboard camera, we use a CNN to infer projections of the skeletal model, enabling correspondence constraints to be established through a cost function. This is then coupled with GPS/IMU measurements taken at key frames in the graph to allow successive optimisation as the drone navigates around the turbine. We present two variants of the cost function, one based on traditional 2D point correspondences and the other on direct image interpolation within the inferred projections. Results from experiments on simulated and real-world data show that simultaneous optimisation provides improvements to localisation over only optimising the pose and that combined use of both cost functions proves most effective.

Reasoning about spatial and geometric relations between objects in a tabletop human-robot interaction is a challenge due to the perception not being always consistent: objects placed on a table seem to be slightly in the air; they overlap; they disappear due to occlusions. Yet, interpreting and anchoring perceptual data in a physically consistent estimation of the scene is a crucial ability for humans, and thus robots in HRI context. In this paper we present a simulation-based physics reasoner integrated in a lightweight situation-assessment framework called Underworlds, that allows the robot to stabilize objects and build at run-time a consistent estimation of the scene, even for entirely hidden objects, while inferring the actions performed by its human partner.

We present Simitate - a hybrid benchmarking suite targeting the evaluation of approaches for imitation learning. A dataset containing 1938 sequences where humans perform daily activities in a realistic environment is presented. The dataset is strongly coupled with an integration into a simulator. RGB and depth streams with a resolution of 960×540 at 30Hz and accurate ground truth poses for the demonstrator's hand, as well as the object in 6 DOF at 120Hz are provided. Along with our dataset we provide the 3D model of the used environment, labeled object images and pre-trained models. A benchmarking suite that aims at fostering comparability and reproducibility supports the development of imitation learning approaches. Further, we propose and integrate evaluation metrics on assessing the quality of effect and trajectory of the imitation performed in simulation. Simitate is available on our project website: https://agas.uni-koblenz.de/simitate/.

We present a new approach for transfer of dynamic robot control policies such as biped locomotion from simulation to real hardware. Key to our approach is to perform system identification of the model parameters μ of the hardware (e.g. friction, center-of-mass) in two distinct stages, before policy learning (pre-sysID) and after policy learning (post-sysID). Pre-sysID begins by collecting trajectories from the physical hardware based on a set of generic motion sequences. Because the trajectories may not be related to the task of interest, presysID does not attempt to accurately identify the true value of μ, but only to approximate the range of μ to guide the policy learning. Next, a Projected Universal Policy (PUP) is created by simultaneously training a network that projects μ to a low-dimensional latent variable η and a family of policies that are conditioned on η. The second round of system identification (post-sysID) is then carried out by deploying the PUP on the robot hardware using task-relevant trajectories. We use Bayesian Optimization to determine the values for η that optimize the performance of PUP on the real hardware. We have used this approach to create three successful biped locomotion controllers (walk forward, walk backwards, walk sideways) on the Darwin OP2 robot.

This paper addresses the problem of human body detection-particularly a human body lying on the ground (a.k.a. casualty)-using point cloud data. This ability to detect a casualty is one of the most important features of mobile rescue robots, in order for them to be able to operate autonomously. We propose a deep-learning-based casualty detection method using a deep convolutional neural network (CNN). This network is trained to be able to detect a casualty using a point-cloud data input. In the method we propose, the point cloud input is pre-processed to generate a depth image-like ground-projected heightmap. This heightmap is generated based on the projected distance of each point onto the detected ground plane within the point cloud data. The generated heightmap-in image form-is then used as an input for the CNN to detect a human body lying on the ground. To train the neural network, we propose a novel sim-to-real approach, in which the network model is trained using synthetic data obtained in simulation and then tested on real sensor data. To make the model transferable to real data implementations, during the training we adopt specific data augmentation strategies with the synthetic training data. The experimental results show that data augmentation introduced during the training process is essential for improving the performance of the trained model on real data. More specifically, the results demonstrate that the data augmentations on raw point-cloud data have contributed to a considerable improvement of the trained model performance.

Quadrotor stabilizing controllers often require careful, model-specific tuning for safe operation. We use reinforcement learning to train policies in simulation that transfer remarkably well to multiple different physical quadrotors. Our policies are low-level, i.e., we map the rotorcrafts' state directly to the motor outputs. The trained control policies are very robust to external disturbances and can withstand harsh initial conditions such as throws. We show how different training methodologies (change of the cost function, modeling of noise, use of domain randomization) might affect flight performance. To the best of our knowledge, this is the first work that demonstrates that a simple neural network can learn a robust stabilizing low-level quadrotor controller (without the use of a stabilizing PD controller) that is shown to generalize to multiple quadrotors. The video of our experiments can be found at https://sites.google.com/view/sim-to-multi-quad.

Visual Servoing (VS), where images taken from a camera typically attached to the robot end-effector are used to guide the robot motions, is an important technique to tackle robotic tasks that require a high level of accuracy. We propose a new neural network, based on a Siamese architecture, for highly accurate camera pose estimation. This, in turn, can be used as a final refinement step following a coarse VS or, if applied in an iterative manner, as a standalone VS on its own. The key feature of our neural network is that it outputs the relative pose between any pair of images, and does so with sub-millimeter accuracy. We show that our network can reduce pose estimation errors to 0.6 mm in translation and 0.4 degrees in rotation, from initial errors of 10 mm / 5 degrees if applied once, or of several cm / tens of degrees if applied iteratively. The network can generalize to similar objects, is robust against changing lighting conditions, and to partial occlusions (when used iteratively). The high accuracy achieved enables tackling low-tolerance assembly tasks downstream: using our network, an industrial robot can achieve 97.5% success rate on a VGA-connector insertion task without any force sensing mechanism.

Current methods for autonomous management use strict first-come, first-serve (FCFS) ordering to manage incoming autonomous vehicles at an intersection. In this work, we present a coordination policy that swaps agent ordering to increase the system-wide performance while ensuring that the swaps are socially compliant. By considering an agent's Social Value Orientation (SVO), a social psychology metric for their willingness to help another vehicle, the central coordinator can reduce system delays while ensuring each individual vehicle increases their own utility. The FCFS-SVO algorithm is both computationally tractable and accounts for a variety of real-world agent types, such as human drivers and a variety of social orientations. Simulation results show that average vehicle delays decrease with swapping by enabling cooperation between agents. In addition, we show that the proportion of human drivers, as well as, the distribution of prosocial and egoistic vehicles in the system can have a prominent effect on the performance of the system.

During robot-assisted surgery, a Remote Center of Motion (RCM) is often implemented to constrain instrument motion through and about a specific point in space. Aligning and re-positioning this point during surgery is not trivial, as this is a defined, yet often non-visualised, point in space. When misaligned with the patient surrounding anatomy is excessively strained, potentially causing post-operative complications. Not being able to safely re-position the RCM for these purposes limits the use of surgical robotics. This work introduces a general approach relying on anatomy-based haptic fixtures to simplify and improve RCM positioning during robot-assisted surgery. The proposed approach is extended with a novel method and mechanism to mechanically implement such fixtures. This is applied for the use case of vitreoretinal surgery, for which purpose a dedicated robotic setup and fixture mechanism was developed. These were used to conduct an initial experimental validation, during which the feasibility of both a virtual and mechanical implementation is reviewed. Initial outcomes suggest that both implementations are feasible. With the currently used impedance-type system, the mechanical implementation is shown to offer at least one order of magnitude stiffness increase when compared to an equivalent virtual implementation.

The sense of touch is often crucial for humans to perform manipulation tasks. Providing tactile feedback during teleoperation or for users of prosthetic devices would be beneficial. However, the representation of tactile information constitutes a major technical challenge, since the numerous and possibly multimodal sensor readings are massive compared to the available tactile display technology. We introduce an algorithm that deploys two stages of K-means clustering along and across tactile image frames that render tactile sensor information at each time instant. In this manner, the massive tactile information is adaptively compressed in real-time while preserving its physical meaning, thus, remains intuitive and direct. We experimentally verify and examine the characteristics of our algorithm by evaluating the original and compressed tactile data. The data was gathered during the active tactile exploration of several objects of daily living by an Allegro robot hand that was covered with 15 uSkin sensor modules providing 2403-axis force vector measurements at each time instant. Our novel algorithm is straight forward enough to be implemented into tactile feedback systems. Finally, our algorithm allows for the direct feedback of massive tactile sensor data for a broad variety of tactile sensors and tactile displays, thereby, enables the compressed yet intuitive representation of massive tactile sensor information for real-time applications.

Place recognition and loop-closure detection are main challenges in the localization, mapping and navigation tasks of self-driving vehicles. In this paper, we solve the loop-closure detection problem by incorporating the deep-learning based point cloud description method and the coarse-to-fine sequence matching strategy. More specifically, we propose a deep neural network to extract a global descriptor from the original large-scale 3D point cloud, then based on which, a typical place analysis approach is presented to investigate the feature space distribution of the global descriptors and select several super keyframes. Finally, a coarse-to-fine strategy, which includes a super keyframe based coarse matching stage and a local sequence matching stage, is presented to ensure the loop-closure detection accuracy and real-time performance simultaneously. Thanks to the sequence matching operation, the proposed approach obtains an improvement against the existing deep-learning based methods. Experiment results on a self-driving vehicle validate the effectiveness of the proposed loop-closure detection algorithm.