IROS 2018 Papers

Neuroplastic changes in motor cortex is essential for the recovery motor function of patients with neurological diseases. To enlarge neuroplastic change, various movements should be provided to stimulate larger motor cortical area, and because hands occupy the largest area, it is especially important. Many wearable robotic devices have been developed for rehabilitation of the hand, and soft robotic devices in particular have drawn attention for their compact design. However, most soft devices provide simple thumb motions, which flex or extend all joints without assistance of opposition/reposition of the carpometacarpal joint although the importance in producing various grasps. In this study, the design of a cable actuated dexterous (CADEX) glove is proposed. For dexterous motion, the structure and orientation of major finger tendons were replicated with exotendons (actuated cables), and four exotendons were used for the thumb with the path optimized to provide flexion/extension of the thumb and decoupled opposition/reposition of the carpometacarpal with other joints. To provide consistent motion, silicon was used for stable anchoring of exotendons while preventing slippage and reducing deformation. The motion generated by the CADEX glove was experimentally evaluated for a single healthy subject. The result shows that the CADEX glove could flex and extend the finger with various ratios among joints, and the opposition/reposition of carpometacarpal joint of the thumb could be achieved consistently with minimal effect on the other joints. The CADEX glove is expected to help providing various tasks which is expected to enhance the functional recovery of patients with neurological disease.

Since the resurgence of CNNs the robotic vision community has developed a range of algorithms that perform classification, semantic segmentation and structure prediction (depths, normals, surface curvature) using neural networks. While some of these models achieve state-of-the art results and super human level performance, deploying these models in a time critical robotic environment remains an ongoing challenge. Real-time frameworks are of paramount importance to build a robotic society where humans and robots integrate seamlessly. To this end, we present a novel real-time structure prediction framework that predicts depth at 30 frames per second on an NVIDIA-TX2. At the time of writing, this is the first piece of work to showcase such a capability on a mobile platform. We also demonstrate with extensive experiments that neural networks with very large model capacities can be leveraged in order to train accurate condensed model architectures in a “from teacher to student” style knowledge transfer.

We tackle the transition feasibility problem, that is the issue of determining whether there exists a feasible motion connecting two configurations of a legged robot. To achieve this we introduce CROC, a novel method for computing centroidal dynamics trajectories in multi-contact planning contexts. Our approach is based on a conservative and convex reformulation of the problem, where we represent the center of mass trajectory as a Bezier curve comprising a single free control point as a variable. Under this formulation, the transition problem is solved efficiently with a Linear Program (LP)of low dimension. We use this LP as a feasibility criterion, incorporated in a sampling-based contact planner, to discard efficiently unfeasible contact plans. We are thus able to produce robust contact sequences, likely to define feasible motion synthesis problems. We illustrate this application on various multi-contact scenarios featuring HRP2 and HyQ. We also show that we can use CROC to compute valuable initial guesses, used to warm-start non-linear solvers for motion generation methods. This method could also be used for the 0 and 1-Step capturability problem. The source code of CROC is available under an open source BSD-2 License.

Complex tasks require the combination of both discrete and rhythmic movements. Though scientists do not yet agree on the neural architecture involved in both types and in the transition from one to the other, the importance of having robot controllers able to behave rhythmically and discretely is universally recoanized. In this paper, a bio-inspired robot controller based on oscillating neurons is proposed to realize both discrete and rhythmic movements and easily transition from one to the other. It is shown that, under certain parameter conditions, the CPG controller behaves like a PID controller. In order to demonstrate the feasibility of controlling both discrete and rhythmic movements, the CPG is applied to the initiation of handshaking, namely, reach towards the human hand and start to shake it. Results show that this architecture is suitable for both discrete and rhythmic movements and can easily transition from one to the other.

This paper presents the first two members of the new generation of CLASH hands, which exploit low cost actuation and rapid prototyping to create antagonistic modular and lightweight hands and grippers. The hands approach the robustness of the DLR Awiwi hand with a much lower complexity and cost. To reduce the number of required actuators, a differential coupling mechanism for underactuated fingers was developed, along with a new mechanism that uses variable stiffness actuation in order to increase the workspace of underactuated fingers. The hands provide a research platform for both hand-in-hand and robotic grasping. Design aspects are discussed, and an initial experimental validation verifies the hands' performance.

There have been several attempts at modeling context in robots. However, either these attempts assume a fixed number of contexts or use a rule-based approach to determine when to increment the number of contexts. In this paper, we pose the task of when to increment as a learning problem, which we solve using a Recurrent Neural Network. We show that the network successfully (with 98% testing accuracy) learns to predict when to increment, and demonstrate, in a scene modeling problem (where the correct number of contexts is not known), that the robot increments the number of contexts in an expected manner (i.e., the entropy of the system is reduced). We also present how the incremental model can be used for various scene reasoning tasks.

In many applications, maintaining a consistent dense map of the environment is key to enabling robotic platforms to perform higher level decision making. Several works have addressed the challenge of creating precise dense 3D maps from visual sensors providing depth information. However, during operation over longer missions, reconstructions can easily become inconsistent due to accumulated camera tracking error and delayed loop closure. Without explicitly addressing the problem of map consistency, recovery from such distortions tends to be difficult. We present a novel system for dense 3D mapping which addresses the challenge of building consistent maps while dealing with scalability. Central to our approach is the representation of the environment as a collection of overlapping Truncated Signed Distance Field (TSDF) subvolumes. These subvolumes are localized through feature-based camera tracking and bundle adjustment. Our main contribution is a pipeline for identifying stable regions in the map, and to fuse the contributing subvolumes. This approach allows us to reduce map growth while still maintaining consistency. We demonstrate the proposed system on a publicly available dataset and simulation engine, and demonstrate the efficacy of the proposed approach for building consistent and scalable maps. Finally we demonstrate our approach running in real-time onboard a lightweight Micro Aerial Vehicle (MAV).

Risk-aware Multi-Policy Decision Making (MPDM)is a powerful framework for reliable navigation in a dynamic social environment where rather than evaluating individual trajectories, a “library” of policies (reactive controllers)is evaluated by anticipating potentially dangerous future outcomes using an on-line forward roll-out process. There is a core tension in Multi-Policy Decision Making (MPDM)systems - it is desirable to add more policies to the system for flexibility in finding good policies, however, this increases computational cost. As a result, MPDM was limited to small (perhaps 5-10)discrete policies - a significant performance bottleneck. In this paper, we radically enhance the expressivity of MPDM by allowing policies to have continuous-valued parameters, while simultaneously satisfying real-time constraints by quickly discovering promising policy parameters through a novel iterative gradient-based algorithm. Our evaluation includes results from extensive simulation and real-world experiments in semi-crowded environments.

The ability to build models of the environment is an essential prerequisite for many robotic applications. In recent years, mapping of dense surface geometry using RGB-D cameras has seen extensive progress. Many approaches build colored models, typically directly using the intensity values provided by the camera. Unfortunately, these intensities are inherently affected by illumination. Therefore, the resulting maps only represent the environment for one specific lighting condition. To overcome this limitation, we propose to build reflectance maps that are invariant against changes in lighting. Our approach estimates the diffuse reflectance of a surface by recovering its radiosity and the corresponding irradiance. As imperfections in this process can significantly degrade the reflectance estimate, we remove outliers in the high dynamic range radiosity estimation and propose a method to refine the reflectance estimate. Our system implements the whole pipeline for offline reconstruction of dense reflectance maps including the segmentation of light emitters in the scene. We demonstrate the applicability of our approach in real-world experiments under varying lighting conditions.

This paper presents a body-mounted, four degree-of-freedom (4-DOF) parallel mechanism robot for image-guided percutaneous interventions. The design of the robot is optimized to be light weight and compact such that it could be mounted to the patient body. It has a modular design that can be adopted for assisting various image-guided, needle-based percutaneous interventions such as arthrography, biopsy and brachytherapy seed placement. The robot mechanism and the control system are designed and manufactured with components compatible with imaging modalities including Magnetic Resonance Imaging (MRI) and Computed Tomography (CT). The current version of the robot presented in this paper is optimized for shoulder arthrography under MRI guidance; a Z-shaped fiducial frame is attached to the robot, providing accurate and repeatable robot registration with the MR scanner coordinate system. Here we present the mechanical design of the manipulator, robot kinematics, robot calibration procedure, and preliminary bench-top accuracy assessment. The bench-top accuracy evaluation of the robotic manipulator shows average translational error of 1.01 mm and 0.96 mm in X and Z axes, respectively, and average rotational error of 3.06 degrees and 2.07 degrees about the X and Z axes, respectively.

For unmanned rescue, observation, and/or research, vehicles with high terrain adaptability, high speed, and high reliability are needed to reach hard-to-reach-locations. In order to extend the areas that can be explored, we propose a method and a robot capable of running on the surface of water without having to bypass the puddles and streams that exist on uneven terrain. The method that enables the robot to run on the water surface is bio-inspired by the basilisk lizard that can walk on the surface of water. We developed a blade-type crawler robot with a simple and reliable mechanism, capable of traversing uneven terrain at high speed. The robot with the method was tested on a real water surface and the result confirmed the ability of the robot to run on the water surface.

This paper presents the design and validation of controlling hopping on the 3D bipedal robot Cassie. A spring-mass model is identified from the kinematics and compliance of the robot. The spring stiffness and damping are encapsulated by the leg length, thus actuating the leg length can create and control hopping behaviors. Trajectory optimization via direct collocation is performed on the spring-mass model to plan jumping and landing motions. The leg length trajectories are utilized as desired outputs to synthesize a control Lyapunov function based quadratic program (CLF-QP). Centroidal angular momentum, taking as an addition output in the CLF-QP, is also stabilized in the jumping phase to prevent whole body rotation in the underactuated flight phase. The solution to the CLF-QP is a nonlinear feedback control law that achieves dynamic jumping behaviors on bipedal robots with compliance. The framework presented in this paper is verified experimentally on the bipedal robot Cassie.

Versatile multi-modal robots are advantageous for their wider operational environments. By taking design principles from observation of Pteromyini, commonly known as the flying squirrel, which shows balanced performances in both aerial and terrestrial locomotion, a novel robotic platform with the ability of gliding and walking is designed. The flexible membrane and gliding method of Pteromyini have been applied to the robot design. The legs of the robot were optimized to perform with regulated motor torques in both walking and gliding. The robot glided with an average gliding ratio of 1.88 and controlled its angle-of-attack for slowing down to land safely. The robot was able to walk utilizing different gait patterns. These results demonstrated our robot's balanced multi-modal locomotion of gliding and walking.

Bimanual assembly of two parts require that a relative target pose is reached prior to the joining operation. Rather than utilizing one arm as a fixture for holding one of the parts while the other performs the assembly, motion generation in the relative end-effector frame is proposed that involves both arms. The proposed approach considers bimanual motion in a dynamic and uncertain environment addressing avoidance of collision with obstacles as well as the robot itself and the environment. Moreover, configurations that optimize the motion and force capabilities for the sucessful and efficient completion of the task are taken into account. A task priority strategy is adopted achieving online performance. Experimental results on the YuMi bimanual robot using the Stack-Of- Tasks hierarchical solver validate the performance of the proposed approach in a folding assembly task.

This paper is devoted to the study of multirotor Micro Aerial Vehicles (MAVs) with fixed-pitch propellers and bidirectional thrust vector. The latter is realized by using dedicated motor controllers, which allow to invert the propellers' direction of rotation during flight (so-called 3D mode), and almost or fully symmetric propellers. We present a unified modeling, controller design, and control allocation approach that accounts for bidirectional thrust. Suitable propellers with the ability to produce thrust and torque in both directions are compared and their parameters are identified through a static thrust test. Furthermore, we discuss applications of bidirectional thrust, like inverted flight or surface slip reduction, which are impossible to realize with common unidirectional thrust vehicles. We generate suitable flight trajectories and evaluate our unified approach in experiments with a custom-built quadrotor.

This paper presents a novel method for generating the trajectory of a robot assisting a human in servicing a set of tasks embedded in a convex 2-D domain. This method makes use of Bayesian inference to predict human intent in task selection. Rather than following optimal trajectory towards a single task, the robot computes a set of potentially optimal tasks each weighted by the human's posterior probability and superimposes them into a cost function that is designed to minimize the weighted Euclidean distance relative to set. The effect is a flexible path human-robot collaborative network that is shown in simulation to complete all tasks in a given domain in less time than existing methods for a certain class of highly impulsive humans, i.e., humans that tend to randomly switch tasks at times generated by a Poisson counting process. The algorithm is also illustrated through an experimental demonstration.

Typical inference approaches that work with high-dimensional visual measurements use hand-engineered image features (e.g, SIFT) that require combinatorial data association, or predict only hidden state mean without considering its uncertainty and multi-modality aspects. We develop a novel approach to infer system hidden state from visual observations via CNN features which are outputs of a CNN classifier. To that end, at pre-deployment stage we use neural networks to learn a generative viewpoint-dependent model of CNN features given the robot pose and approximate this model by a spatially-varying Gaussian distribution. Further, at deployment this model is utilized within a Bayesian framework for probabilistic inference, considering a robot localization problem. Our method does not involve data association and provides uncertainty covariance of the final estimation. Moreover, we show empirically that the CNN feature likelihood is unimodal which simplifies the inference task. We test our method in a simulated Unreal Engine environment, where we succeed to retrieve high-level state information from CNN features and produce trajectory estimation with high accuracy. Additionally, we analyze robustness of our approach to different light conditions.

It has been observed that a robot shown as suffering is enough to cause an empathic response from a person. Whether the response is a fleeting reaction with no consequences or a meaningful perspective change with associated behavior modifications is not clear. Existing work has been limited to measurements made at the end of empathy inducing experimental trials rather measurements made over time to capture consequential behavioral pattern. We report on preliminary results collected from a study that attempts to measure how the actions of a participant may be altered by empathy for a robot companion. Our findings suggest that induced empathy can in fact have a significant impact on a person's behavior to the extent that the ability to fulfill a mission may be affected.

Soft actuators with auxetic, or negative Poisson's ratio (NPR), behavior offer a way to create soft robots with novel kinematic behavior. This paper presents an original framework for reinforcement of a soft actuator using a generalized NPR element, called a Representative Auxetic Element (RAE), and an experimental validation of the kinematic behavior that it enables. We build a generalized kinematic model that enables the design of RAE-patterned actuators and reveal the distinct auxetic behavior of RAE actuators with comparable model accuracy to the legacy McKibben actuators. A simple, reproducible way of designing and fabricating RAE actuators is described and varied prototypes are shown. This RAE-based design scheme can be used to create actuators with specified kinematics like bending, extension, and radial expansion, which can also vary across the actuator's surface both circumferentially and axially in a tractable, scalable manner.