IROS 2018 Papers

In this paper, we present a new characterization of mobility for formations of unicycle robots defined by distance-bearing constraints. In fact, by introducing a simple reduction procedure which associates a prescribed formation with a “macro-robot”, we extend the classification by type proposed by Campion et al., to multi-agent systems. To simplify the classification task, which only leverages the nonslip condition for a conventional centered wheel, we assume that the robots are disposed at the vertices of a regular convex polygon. We demonstrate the practical utility of the notion of macro-robot in a trajectory-tracking control problem for a formation of unicycles.

Crossmodal conflict resolution is crucial for robot sensorimotor coupling through the interaction with the environment, yielding swift and robust behaviour also in noisy conditions. In this paper, we propose a neurorobotic experiment in which an iCub robot exhibits human-like responses in a complex crossmodal environment. To better understand how humans deal with multisensory conflicts, we conducted a behavioural study exposing 33 subjects to congruent and incongruent dynamic audio-visual cues. In contrast to previous studies using simplified stimuli, we designed a scenario with four animated avatars and observed that the magnitude and extension of the visual bias are related to the semantics embedded in the scene, i.e., visual cues that are congruent with environmental statistics (moving lips and vocalization) induce the strongest bias. We implement a deep learning model that processes stereophonic sound, facial features, and body motion to trigger a discrete behavioural response. After training the model, we exposed the iCub to the same experimental conditions as the human subjects, showing that the robot can replicate similar responses in real time. Our interdisciplinary work provides important insights into how crossmodal conflict resolution can be modelled in robots and introduces future research directions for the efficient combination of sensory observations with internally generated knowledge and expectations.

Existing adaptive robot control laws typically require an engineering choice of a constant adaptation gain matrix, which often involves repeated and time-consuming trial and error. Moreover, physical consistency of the estimated inertial parameters or the uniform positive definiteness of the estimated robot mass matrix cannot in general be guaranteed without nonsmooth corrections, e.g., projection to the boundary of the feasible parameter set. In this paper we present a natural adaptive control law that mitigates many of these difficulties, by exploiting the coordinate-invariant differential geometric structure of the space of physically consistent inertial parameters. Our approach provides a more generalizable and physically consistent adaptation law for the robot parameters without significant additional computations compared to existing methods. Simulation results showing markedly improved tracking error convergence over existing adaptive control laws are provided as validation.

This paper presents the design, modeling and fabrication of a millimeter-size Continuum Soft Robot (CSR). The robot consists of active flexible polymer actuator-based multisegment robot. A multiphysics model based on multilayer cantilever for large displacement is established between the input voltages to the distal tip position of a single segment robot. The extension of the model to multisegment CSR is derived. The proposed model is validated experimentally then a two-segment CSR and three-segment CSR in 3D arrangement are investigated, demonstrating the model efficiency for obtaining complex configuration. Moreover, various configurations can be explored to derive complex kinematics then increasing the robot capability.

This paper introduces an extension of the reverse priority framework for the kinematic control of redundant robots. It integrates, in a unified framework, the treatment of multiple tasks, multiple kinematic chains, different joint priorities and hard constraints. The management of multiple tasks is based on the reverse priority method, that has been modified so that it makes possible the assignment of different priorities to each joint in order to accomplish the tasks. This framework is also suitable for robotic systems with multiple kinematic chains, which could share several joints. Moreover, it can deal with bilateral and unilateral constraints, that can be defined either at joint or cartesian space. Hard constraints are considered at each priority level, instead of treating them separately at the highest priority level. The proposed framework has been evaluated in simulation and in real experiments with a redundant underwater vehicle-manipulator system at sea.

Minimally invasive surgery can benefit greatly from utilizing micro-agents. These miniaturized agents need to be clearly visualized and precisely controlled to ensure the success of the surgery. Since medical imaging modalities suffer from low acquisition rate, multi-rate sampling methods can be used to estimate the intersample states of micro-agents. Hence, the sampling rate of the controller can be virtually increased even if the position data is acquired using a slow medical imaging modality. This study presents multi-rate Luenberger and Kalman state estimators for visual tracking of micro-agents. The micro-agents are tracked using sum of squared differences and normalized cross correlation based visual tracking. Further, the outputs of the two methods are merged to minimize the tracking error and prevent tracking failures. During the experiments, the micro-agents with different geometrical shapes and sizes are imaged using a 2D ultrasound machine and a microscope, and manipulated using electromagnetic coils. The multi-rate state estimation accuracy is measured using a high speed camera. The precision of the tracking and multi-rate state estimation are verified experimentally under challenging conditions. For this purpose, an elliptical shaped magnetic micro-agent with a length of 48 pixels is used. Maximum absolute error in x and y axes are 2.273 and 2.432 pixels for an 8-fold increase of the sample rate (25 frames per second), respectively. During the experiments, it was observed that the micro-agents could be tracked more reliably using normalized cross correlation based visual tracking and inters ample states could be estimated more accurately using Kalman state estimator. Experimental results show that the proposed method could be used to track micro-agents in medical imaging modalities and estimate system states at intermediate time instants in real-time.

Robust localization is a prerequisite for autonomous vehicles. Traditional visual localization methods like visual odometry suffer error accumulation on long range navigation. In this paper, a flexible road map based probabilistic filtering method is proposed to tackle this problem. To effectively match the ego-trajectory to various curving roads in map, a new representation based on anchor point (AP) which captures the main curving points on the trajectory is presented. Based on APs of the map and trajectory, a flexible Multi-Position Joint Particle Filtering (MPJPF) framework is proposed to correct the position error. The method features the capability of adaptively estimating a series of APs jointly and only updates the estimation at situations with low uncertainty. It explicitly avoids the drawbacks of obliging to determine the current position at large uncertain situations such as dense parallel road branches. The experiments carried out on KITTI benchmark demonstrate our success.

This paper outlines an algorithmic approach for the automatic coordination and planning of heterogeneous multi-robot teams. Specifically, this work addresses the marsupial-based subset of multi-robot teams, where “carrier” robots transport and deploy “passenger” robots. The approach starts with a high-level watershed segmentation of the world to determine the free-space regions accessible by each robot in the team. Topological graph planning then decides the high-level motion plan for each robot between these free-space regions. Finally, a low-level path planner generates optimized, dynamically-feasible trajectories for each robot along the topological path. The performance of the approach is evaluated in simulation and through hardware experiments.

The main contribution of this paper is an extended Kalman filter (EKF)based algorithm for estimating the 6 DOF pose of a camera using monocular images of an indoor environment. In contrast to popular visual simultaneous localisation and mapping algorithms, the technique proposed relies on a pre-built map represented as an unsigned distance function of the ground plane edges. Images from the camera are processed using a Convolutional Neural Network (CNN)to extract a ground plane edge image. Pixels that belong to these edges are used in the observation equation of the EKF to estimate the camera location. Use of the CNN makes it possible to extract ground plane edges under significant changes to scene illumination. The EKF framework lends itself to use of a suitable motion model, fusing information from any other sensors such as wheel encoders or inertial measurement units, if available, and rejecting spurious observations. A series of experiments are presented to demonstrate the effectiveness of the proposed technique.

This paper presents the design of a vision-based object grasping and motion control architecture for a mobile manipulator system. The optimal grasping areas of the object are estimated using the partial point cloud acquired from an onboard RGB-D sensor system. The reach-to-grasp motion of the mobile manipulator is handled via a Nonlinear Model Predictive Control scheme. The controller is formulated accordingly in order to allow the system to operate in a constrained workspace with static obstacles. The goal of the proposed scheme is to guide the robot's end-effector towards the optimal grasping regions with guaranteed input and state constraints such as occlusion and obstacle avoidance, workspace boundaries and field of view constraints. The performance of the proposed strategy is experimentally verified using an 8 Degrees of Freedom KUKA Youbot in different reach-to-grasp scenarios.

A combination of several autonomous UAVs can be used to perform collaborative tasks. Such a combination is referred to as a swarm of drones. The use of multiple platforms can extend the system global capacities thanks to the resulting variety of embedded sensors and to information sharing. In this case, path planning and thus obstacles avoidance is still a major task. To deal with this issue, mobility models have to be implemented. Our contribution presented in this paper is a mobility model for swarms of UAVs based on the Artificial Potential Fields (APF) principle. In our model, the involved UAVs collaborate by sharing data about the obstacles that they detected. By doing so, a UAV which is not close enough to an obstacle to detect it thanks to its own sensors will still have the proper data to take this obstacle into account in its path planning. To validate our mobility strategies with realistic constraints we simulate the performances of existing sensors and transmitters, and consider real-world environment.

Collaborative robots are often designed with limited power and force capacity, with the aim to provide affordable solutions and ensure human safety in case of accidental collisions and impacts. If a task requires a power beyond this capacity, or is performed repeatedly over long periods, such limits may be exceeded, which can cause inevitable robot damage and contribute to the lost productivity. In such cases, where hardware solutions and improvements are not applicable, effective software frameworks can prolong robot productivity and lifetime. To this end, in this paper we propose a novel technique for the monitoring and management of robot fatigue in repetitive or high-effort task execution scenarios. The robot fatigue is estimated by the measured temperature of motors in the joints. The proposed fatigue management system is composed of two-stage reaction process that is triggered by different levels of the estimated fatigue. The first stage exploits the kinematic redundancy of robot structure in attempt to minimise the load in the specific joints that under fatigue by reconfiguration in the joint space through the null space of the Cartesian task production. If the first stage is not successful in reducing the fatigue, the second stage is activated that gradually reduces the forces of hybrid controller. At that point, the human co-worker can temporarily take over the task execution until the robot will be recovered from the excessive fatigue. To validate the proposed approach we conducted experiments on KUKA Lightweight Robot performing two interaction tasks: autonomous surface wiping and collaborative human-robot surface polishing.

Man-made environments such as households, offices, or factory floors are typically composed of linear structures. Accordingly, polylines are a natural way to accurately represent their geometry. In this paper, we propose a novel probabilistic method to extract polylines from raw 2-D laser range scans. The key idea of our approach is to determine a set of polylines that maximizes the likelihood of a given scan. In extensive experiments carried out on publicly available real-world datasets and on simulated laser scans, we demonstrate that our method substantially outperforms existing state-of-the-art approaches in terms of accuracy, while showing comparable computational requirements. Our implementation is available under https://github.com/acschaefer/ple.

For traditional manipulators, there is a large number of electrical cables between the motion controller and the joint servo controllers. It is very inconvenient for maintenance, update, and safe operation. In this paper, we develop a lightweight redundant manipulator with high stable wireless communication link and compliance control. The motion controller, servo controller, and communication link are taken as a whole system to be optimized. The manipulator body and the motion controller are physically separated. It is very helpful for building distributed networked-manufacturing system or intelligent manufacturing system for Industry 4.0. The control system can be quickly updated by changing the object's identification without reconnect the communication cables. The mechanical part of the manipulator contains modular joints and links. Each joint is integrated with hall sensors, an incremental magnetic encoder, an absolute magnetic encoder and current sensors. The electrical part includes a central controller, seven joint servo controllers, and a wireless communication module based on ZigBee. By designing the application layer protocol, the communication stability is improved. In order to achieve the force control requirements in fine operation like assembly. A wireless compliance control frame is then designed. The compliance control method is realized on the central controller, by which the generated control commands are sent to the joint servo controllers through a wireless link. The problems caused by large electrical cables are then solved. Finally, the prototype and the experimental system are developed. Some experiments are carried out, including wireless communication test, trajectory tracking experiments, load carrying experiments, and wireless impedance control experiments. Results verify the functions and performance of the developed 7-DOF manipulator.

Robots need to adapt themselves to various surroundings in order to achieve robust object search and grasp in unknown environments. For this adaptation, robot motions should be implemented as combination of primitive motions which are based on sensor reaction. Among various sensing methods, non contact sensing is required as a means of preventing operation failures such as pushing objects. Especially, proximity sensors have been proved effective in avoiding occlusion problems. In this paper, we first develop a gripper on which proximity sensors are mounted all around, and then calculate distance between the gripper and objects using proposed calibration method. This enables robots to recognize detailed shapes of objects surrounding the gripper. We also propose primitive motions for object search and grasp, and describe the contents of each motion. The motions are based on sensor information obtained from the gripper. We verify the effectiveness of our system through an experiment in which a real robot performs complex tasks by combination of the primitive motions.

Suction is an effective method for picking various objects because it makes trajectory planning and control easy. However, suction has not been used due to misalignment and leakage of suction air when handling a variety of shapes. We therefore develop a hand to handle these characteristics. First, we model the vacuum pump and pad characteristics to allow evaluation of momentum and suction force in the case of leakage. Utilizing this, we select a configuration suitable for the items in the Amazon Robotics Challenge 2017. In addition, we design a mechanism for switching from suction to pinching for grasping items that cannot be sucked. Moreover, robust pinching is made possible by equipping the fingertips with a passive linear motion mechanism. In the Amazon Robotics Challenge 2017, it was shown possible to stably grasp items with irregularities and items with large moments. Furthermore, items that cannot be grasped by suction can also be grasped robustly by switching to the pinching mechanism.

This paper presents a programming by demonstration framework for teaching impedance modulation using human demonstrations. Physiologically, human stiffness and damping are coupled at the muscle level, restricting the ability to modulate impedance according to task demands. Robotic systems often do not have this restriction (stiffness and damping can be varied independently), but the challenge is to devise an appropriate variable impedance profile for a given task. In this paper, the task critical component is first learned for imitation and a robot-specific controller is then blended into the control using the null space. In doing so, the control cheme takes advantage of both human and robot `best practice'. Experimental results on a physical robot suggest an order of magnitude better mean performance, with lower variance, can be achieved using the blended scheme.

Grasp planning is essential for robots to execute dexterous tasks. Solving the optimal grasps for various objects online, however, is challenging due to the heavy computation load during exhaustive sampling, and the difficulties to consider task requirements. This paper proposes a framework to combine analytic approach with learning for efficient grasp generation. The example grasps are taught by human demonstration and mapped to similar objects by a non-rigid transformation. The mapped grasps are evaluated analytically and refined by an orientation search to improve the grasp robustness and robot reachability. The proposed approach is able to plan high-quality grasps, avoid collision, satisfy task requirements, and achieve efficient online planning. The effectiveness of the proposed method is verified by a series of experiments.

This paper presents the foundations of a MATLAB framework for dynamic modeling and simulation of closed kinematic chain (CKC) mechanisms, with a particular focus on implementation with legged locomotive mechanisms. As such, the framework supports both floating-base and fixed-base systems. Through the use of singular perturbation theory, various CKC mechanisms can be modeled so that constraint errors asymptotically converge to zero, thus avoiding the numerical drift that plagues commonly used methods. A functional API and the relevant core commands necessary to construct a model are presented. Two robotic legs incorporating CKC mechanisms are utilized as case studies, and simulations of each leg performing a dynamic monopedal gait are illustrated.