IROS 2018 Papers

An MRI-actuated catheter is a novel robotic catheter system that utilizes the MR scanner for both remote steering and catheter tracking. In order to develop the mathematical model and the planning algorithm of the catheter in parallel to the MR tracking system, an alternative catheter tracking method is needed. This paper presents a catheter tracking algorithm based on the particle filter and the catadioptric camera system. The motion model of the particle filter is based on the quasi-static kinematics of the catheter. The measurement model calculates the weights of the particles according to the normalized crosscorrelation of the segmented image from camera and a virtual rendering of the catheter. The efficacy of the tracking algorithm is demonstrates via experimental results.

A method for recovery from the complete loss of the state estimate is presented for autonomous quadcopters. Given an aerodynamic force model, the only measurements used to reinitialize the state estimate by means of a bank of extended Kalman filters are the angular rate and linear acceleration measurements of an IMU. The method is integrated within a complete recovery logic on a quadcopter platform and experimentally evaluated.

This paper presents a complete motion planning approach for quadruped locomotion across an unknown terrain using a framework based on mixed-integer convex optimization and visual feedback. Vision data is used to find convex polygons in the surrounding environment, which acts as potentially feasible foothold regions. Then, a goal position is initially provided, which the best feasible destination planner uses to solve for an actual feasible goal position based on the extracted polygons. Next, a footstep planner uses the feasible goal position to plan a fixed number of footsteps, which may or may not result in the robot reaching the position. The center of mass (COM) trajectory planner using quadratic programming is extended to solve for a trajectory in 3D space while maintaining convexity, which reduces the computation time, allowing the robot to plan and execute motions online. The suggested method is implemented as a policy rather than a path planner, but its performance as a path planner is also shown. The approach is verified on both simulation and on a physical robot, ALPHRED, walking on various unknown terrains.

Place recognition is one of the major challenges for the LiDAR-based effective localization and mapping task. Traditional methods are usually relying on geometry matching to achieve place recognition, where a global geometry map need to be restored. In this paper, we accomplish the place recognition task based on an end-to-end feature learning framework with the LiDAR inputs. This method consists of two core modules, a dynamic octree mapping module that generates local 2D maps with the consideration of the robot's motion; and an unsupervised place feature learning module which is an improved adversarial feature learning network with additional assistance for the long-term place recognition requirement. More specially, in place feature learning, we present an additional Generative Adversarial Network with a designed Conditional Entropy Reduction module to stabilize the feature learning process in an unsupervised manner. We evaluate the proposed method on the Kitti dataset and North Campus Long-Term LiDAR dataset. Experimental results show that the proposed method outperforms state-of-the-art in place recognition tasks under long-term applications. What's more, the feature size and inference efficiency in the proposed method are applicable in real-time performance on practical robotic platforms.

Latest deep learning methods for object detection provide remarkable performance, but have limits when used in robotic applications. One of the most relevant issues is the long training time, which is due to the large size and imbalance of the associated training sets, characterized by few positive and a large number of negative examples (i.e. background). Proposed approaches are based on end-to-end learning by back-propagation [22] or kernel methods trained with Hard Negatives Mining on top of deep features [8]. These solutions are effective, but prohibitively slow for on-line applications. In this paper we propose a novel pipeline for object detection that overcomes this problem and provides comparable performance, with a 60x training speedup. Our pipeline combines (i) the Region Proposal Network and the deep feature extractor from [22] to efficiently select candidate RoIs and encode them into powerful representations, with (ii) the FALKON [23] algorithm, a novel kernel-based method that allows fast training on large scale problems (millions of points). We address the size and imbalance of training data by exploiting the stochastic subsampling intrinsic into the method and a novel, fast, bootstrapping approach. We assess the effectiveness of the approach on a standard Computer Vision dataset (PASCAL VOC 2007 [5]) and demonstrate its applicability to a real robotic scenario with the iCubWorld Transformations [18] dataset.

Micro-Aerial Vehicles (MAVs) have the advantage of moving freely in 3D space. However, creating compact and sparse map representations that can be efficiently used for planning for such robots is still an open problem. In this paper, we take maps built from noisy sensor data and construct a sparse graph containing topological information that can be used for 3D planning. We use a Euclidean Signed Distance Field, extract a 3D Generalized Voronoi Diagram (GVD), and obtain a thin skeleton diagram representing the topological structure of the environment. We then convert this skeleton diagram into a sparse graph, which we show is resistant to noise and changes in resolution. We demonstrate global planning over this graph, and the orders of magnitude speed-up it offers over other common planning methods. We validate our planning algorithm in real maps built onboard an MAV, using RGB-D sensing.

A new mechanism for efficiently solving the Markov decision processes (MDPs) is proposed in this paper. We introduce the notion of reachability landscape where we use the Mean First Passage Time (MFPT) as a means to characterize the reachability of every state in the state space. We show that such reachability characterization very well assesses the importance of states and thus provides a natural basis for effectively prioritizing states and approximating policies. Built on such a novel observation, we design two new algorithms - Mean First Passage Time based Value Iteration (MFPT-VI) and Mean First Passage Time based Policy Iteration (MFPT-PI) - that have been modified from the state-of-the-art solution methods. To validate our design, we have performed numerical evaluations in robotic decision-making scenarios, by comparing the proposed new methods with corresponding classic baseline mechanisms. The evaluation results showed that MFPT-VI and MFPT-PI have outperformed the state-of-the-art solutions in terms of both practical runtime and number of iterations. Aside from the advantage of fast convergence, this new solution method is intuitively easy to understand and practically simple to implement.

Compressing soft-obstacles secondary to a controlled motion task is common for human beings. While these tasks are nearly trivial for teleoperated robots, they remain a challenging problem in robotic autonomy. Addressing the problem is significant. For example, in Minimally Invasive Surgeries (MISs), safely compressing soft tissues ensures the surgical safety and decreases tissue removal, thus dramatically decreases surgical trauma and operating room time, and leads to improved surgical outcomes. In this work, we define the problem of soft-obstacle avoidance and project the safety motion constraints into the task space and the velocity space. We illustrate the significance of addressing this problem in the robotic surgery scenario. We present a Recurrent Neural Networks (RNNs) based solution, which formulates the problem as an inequality constrained optimization problem and solves it in its dual space. The application of the proposed method was demonstrated in the Raven II surgical robot. Experimental results demonstrated that the proposed method is effective in addressing the soft-obstacle avoidance problem.

Compliant materials used to create soft robots can better replicate biological structures than typical rigid materials. We can look to nature for developing locomotion strategies for these soft-bodied robots. In this work, snakes were used as inspiration to create an inextensible, soft robot which was used as a platform to test gaits in terrain composed of granular media ranging from fine sand to stone. Snakes vary the speed and amplitude of the traveling wave used in lateral undulation to navigate different environments. We used these gait parameters to develop and test a set of custom gaits that varied the phase offset of the sequence of waves as well as using the best performing gait to test how the amplitude of the wave effects locomotion over the selected terrains. These tests provide preliminary evidence that altering these parameters effects the robot's ability to traverse different terrains. The developed robot is also tested in environments specific to applications for snake robots to show how a soft snake robot can be potentially more effective in these environment. The highest performing gait-curvature combination was the half-activation gait (where the back actuator was activated half as long as the front)with a 135° swept angle. It reached a velocity of 2.2 mm/s or 0.011 body-lengths/s on paper, which was the best performing terrain.

Mobile robots of all shapes and sizes move through the air, water, and over ground. However, few robots can move through the ground. Not only are the forces resisting movement much greater than in air or water, but the interaction forces are more complicated. Here we propose a soft robotic device that burrows through dry sand while requiring an order of magnitude less force than a similarly sized intruding body. The device leverages the principles of both tip-extension and granular fluidization. Like roots, the device extends from its tip; the principle of tip-extension eliminates skin drag on the sides of the body, because the body is stationary with respect to the medium. We implement this with an everting, pressure-driven thin film body. The second principle, granular fluidization, enables a granular medium to adopt a dynamic fluid-like state when pressurized fluid is passed through it, reducing the forces acting on an object moving through it. We realize granular fluidization with a flow of air through the core of the body that mixes with the medium at the tip. The proposed device could lead to applications such as search and rescue in mudslides or shallow subterranean exploration. Further, because it creates a physical conduit with its body, electrical lines, fluids, or even tools could be passed through this channel.

This paper introduces soft LEGO for bottom-up design platform of soft robotics that can be used for various purposes, ranging from research and fast prototyping of soft robots to toys and entertainment. We integrated the interlocking mechanism of LEGO into a modular soft robot. With this design, soft robots could be built by a simple and play-like assembling process. Three kinds of components were proposed to make soft robotics compatible with LEGO: pneumatically inflatable soft brick, flexible bending brick, and channel brick. The soft brick has an air chamber and can generate motions when inflated. The bending brick has flexure and is bendable for generating motion when the assembled soft bricks are pneumatically actuated. The air channel brick has an air channel inside and works as an interface between air hoses and soft LEGO bricks. Detailed design parameters of the soft brick were optimized based on the Taguchi method with finite-element analysis to improve robustness. Design of the bending brick was selected based on experimental results to enhance the robustness of the flexure. Thanks to the multi-material 3-dimensional printer, the soft LEGO bricks could be fabricated with a single printing process. To see the feasibility of soft LEGO as a bottom-up design platform, a simple toy robot for children and a gripper that had a hybrid mechanism of hard and soft materials were built and tested. We hope this soft LEGO could lower the hurdle of soft robotics for children, researchers from other fields, and the public interest in robotics.

This paper presents a fabric actuator consisting of ordinary polymer fibers, conductive fibers, and twisted and coiled soft actuators (TCAs). Previous studies have developed a Spandex TCA (STCA) that is driven at a lower temperature than the conventional Nylon TCA and exhibits greater actuation strain. However, no method to drive STCAs via electrical joule-heating has been developed yet. The fabric actuator presented in this paper offers a solution to this problem by employing an STCA multiple fabrication method, a continuous fabrication method, bundling technology, and weaving technology. Two types of samples (cylindrical and planar) are fabricated and their performances are evaluated experimentally. From the actuation test according to the loads, the maximum contraction strain of 34.3% is measured. The repeatability is also verified through 200 cycles of actuation. Using a linearized model, the dynamic performance of the fabric actuator is predicted and compared with experimental results. An actual human arm size mannequin is driven by applying the fabric actuator, and angle control can be achieved with an encoder mounted on the joint. In addition, fabric actuator is weaved to sweater showing the possibility of wearable assistive robot.

In this work, we show that sensors based on soft, intentionally-lossy optical waveguides are well-suited for soft robotic grasping applications in the deep-sea. Each finger of a soft robotic hand is outfitted with a 2×1 array of optical sensing elements to enable proprioception and contact force sensing. Curvature sensing elements are integrated directly into the structure of a finger, while contact force sensors are fabricated as standalone units and attached afterward. Along with considerations for interfacing with deep-sea remotely operated vehicles (ROVs), models for the effect of bending on light loss and the effect of normal force on strain were used to inform sensor design decisions. Our sensors show sensitivity to curvature over a range of diameters from 8 mm to 76 mm, and sub-Newton force sensitivity. Additionally, sensors were characterized in simulated deep-sea environments at temperatures from -10°C to 50°C and hydrostatic pressures up to 4000 psi. The sensitivity of our curvature sensors is invariant to the temperatures and pressure ranges tested, though contact force sensors decreased in sensitivity as temperatures decreased. Finally, we successfully demonstrate that sensors onboard soft finger actuators can provide informative state feedback during grasping operations in air and water.

The human hand is a complex structure. It is strong but precise. It consists of a very complex mechanical structure that enables the hand to adapt and efficiently handle objects of various shapes, weights and textures. Today's prosthetic devices, struggling to provide similar functions, become overly complex and expensive. They are composed of multiple, precise parts, including miniaturised actuators and sensors as well as complex control, to satisfy the manipulation tasks required. In this paper we propose a soft pneumatic hand that adapts passively to the handled object due to its mechanical compliance. It is pressure driven and enables individual fingers to be controlled independently for dexterity or in groups when a synergistic finger movement is needed. The hand has a truly anatomical shape, is easy to replace and cheap in production. The design can be easily adjusted in terms of shape and size in order to fit each individual user. The paper presents the design, manufacturing technology, current control system and preliminary tests of the hand's capabilities.

For socially assistive robots (SAR)to be accepted into complex and stochastic human environments, it is important to account for subtle social norms. In this paper, we propose a novel approach to socially-aware navigation (SAN)which garnered an immense interest in the Human-Robot Interaction (HRI)community. We use a multi-objective optimization tool called the Pareto Concavity Elimination Transformation (PaC-cET)to capture the non-linear human navigation behavior, a novel contribution to the community. A candidate point on a trajectory is scored (1)for its progress towards the goal, and (2)based on autonomously-sensed distance-based features that capture the social norms and associated social costs. Rather than use a finely-tuned linear combination of these costs, we use PaCcET to select an optimized future trajectory point, associated with a non-linear combination of the costs. Existing research in this domain concentrates on geometric reasoning, model-based, and learning approaches, which have their own pros and cons. This approach is distinct from prior work in this area. We showed in a simulation that the PaCcET-based trajectory planner not only is able to avoid collisions and reach the intended destination in static and dynamic environments but also considers a human's personal space i.e. rules of proxemics in the trajectory selection process.

The integration of migrants and refugees is currently a severe challenge for European states. Especially the imparting of culture- and gender-specific behaviours is an important issue. Social robots might be a valuable tool to introduce refugees to culture-specific behaviours of their host country. In this paper, we investigate the general acceptance of a social robot as well as users' perception of a robot presenting stereo-typical Arabic vs. German female non-verbal behaviour to Syrian newcomers to Germany. Our preliminary study revealed a generally positive attitude towards robots and the idea of an educational robot. Culture-specific manipulations were reflected in participants' partial preference for the Arabic version, but not in participants' perceptual ratings.

People engage in social coordination without explicitly communicating when they are conflicting over spatial resources, e.g., a shop clerk who yields to customers the best place to view products. In this study, we proposed a method that achieves such social coordination with a robot. Our idea is that the social coordination between two agents can be represented as utility-maximizing behavior for joint utility rather than just by a single agent utility. That is, given that each agent's reasonable behavior can be represented as utility-maximizing behavior for single agent utility, we model each agent's plans for himself as well as for the partner agent. Moreover, superiority relationships exist in this joint-utility computation. Since each agent knows such superiority relationships, social coordination can be modeled as utility-yielding behavior based on informed superiority. We specifically focus on looking-together situations for which we developed a utility model. With simulations, we investigate whether the above joint-utility-based modeling successfully reproduces social coordination in looking-together situations. We conducted an experiment in a situation where a tele-operated robot and a customer together look at products in a shop environment. Our experimental results show that our proposed method enables the robot to socially coordinate spatial resources, yielding significantly more thoughtful, less-self-centered, and appropriate impressions than the alternate robot.

Autonomous cars can perform poorly for many reasons. They may have perception issues, incorrect dynamics models, be unaware of obscure rules of human traffic systems, or follow certain rules too conservatively. Regardless of the exact failure mode of the car, often human drivers around the car are behaving correctly. For example, even if the car does not know that it should pull over when an ambulance races by, other humans on the road will know and will pull over. We propose to make socially cohesive cars that leverage the behavior of nearby human drivers to act in ways that are safer and more socially acceptable. The simple intuition behind our algorithm is that if all the humans are consistently behaving in a particular way, then the autonomous car probably should too. We analyze the performance of our algorithm in a variety of scenarios and conduct a user study to assess people's attitudes towards socially cohesive cars. We find that people are surprisingly tolerant of mistakes that cohesive cars might make in order to get the benefits of driving in a car with a safer, or even just more socially acceptable behavior.

Despite being one of the most robust methods in bilateral teleoperation, Time Domain Passivity Approach (TDPA)presents the drawback of accumulating position drift between master and slave devices. The lack of position synchronization poses an obstacle to the performance of teleoperation and may prevent the successful accomplishment of such tasks. Several techniques have been developed in order to solve the position-drift problem in TDPA-based teleoperation. However, they either present poor transparency by over-conservatively constraining force feedback or add high impulse-like force signals that can be harmful to the hardware and to the human operator. We propose a new approach to compensate position drift in TDPA-based teleoperation in a smoother way, which keeps the forces within the normal range of the teleoperation task while preserving the level of transparency and the robust stability of energy-based TDPA. We also add a way of tuning the compensator to behave in accordance with the task being performed, whether it requires faster or smoother compensation. The feasibility and performance of the method were experimentally validated. Good position tracking and regular-amplitude forces are demonstrated with up to 500 ms round-trip constant and variable delays for hard-wall contacts.