IROS 2019 论文列表

Exoskeletons are considered to be a promising technology for assisting and augmenting human performance. A number of challenges related to design, intuitive control and interfaces to the human body must be addressed. In this paper, we approach the question of a minimal sensor setup for the realization of control strategies which take into account the actions currently performed by the user. To this end, we extend our previous work on online classifications of a human wearing a lower limb exoskeleton in two directions. First, we investigate the minimal number of sensors that should be attached to the exoskeleton to achieve a certain classification accuracy by investigating different sensor setups. We compare results of motion classification of 14 different daily activities such as walking forward and going upstairs using Hidden Markov Models. Second, we analyse the influence of different window sizes, as well as the classification performance of different motion types when training on multi- and single-subjects. Our results reveal that we can reduce our sensor setup significantly while achieving about the same classification performance.

Pneumatic rotational stepper motors can be used to actuate MR (magnetic resonance) safe robotic systems. This paper describes novel techniques to minimize the volumetric size and/or step size of such motors in order to cope with the limited space requirements while still delivering high precision. Three designs are presented: the R-10 measures 1.0 cm3, has step size 12.9° and torque 1.2 N mm. The R-40 measures 25.6 cm3, has step size 1.01° and torque 470 N mm. The R-54 measures 46.7 cm3, has step size 1.01 m° and torque 240 N mm. The particularly small step size in the R-54 motor is achieved by using a high-reduction planetary gear.These three motors demonstrate that small-scale rotational stepper motors with a wide range of step sizes and good torque characteristics can be constructed that surpass state-of-art designs by a considerable margin. This allows the advancement of MR safe robotics towards more compact and versatile designs, and also overcome certain existing limitations by combining multiple motors with different specifications.

Metric localization plays a critical role in vision-based navigation. For overcoming the degradation of matching photometry under appearance changes, recent research resorted to introducing geometry constraints of the prior scene structure. In this paper, we present a metric localization method for the monocular camera, using the Signed Distance Field (SDF) as a global map representation. Leveraging the volumetric distance information from SDFs, we aim to relax the assumption of an accurate structure from the local Bundle Adjustment (BA) in previous methods. By tightly coupling the distance factor with temporal visual constraints, our system corrects the odometry drift and jointly optimizes global camera poses with the local structure. We validate the proposed approach on both indoor and outdoor public datasets. Compared to the state-of-the-art methods, it achieves a comparable performance with a minimal sensor configuration.

Multi-objective reinforcement learning (MORL) is the generalization of standard reinforcement learning (RL) approaches to solve sequential decision making problems that consist of several, possibly conflicting, objectives. Generally, in such formulations, there is no single optimal policy which optimizes all the objectives simultaneously, and instead, a number of policies has to be found each optimizing a preference of the objectives. In this paper, we introduce a novel MORL approach by training a meta-policy, a policy simultaneously trained with multiple tasks sampled from a task distribution, for a number of randomly sampled Markov decision processes (MDPs). In other words, the MORL is framed as a meta-learning problem, with the task distribution given by a distribution over the preferences. We demonstrate that such a formulation results in a better approximation of the Pareto optimal solutions in terms of both the optimality and the computational efficiency. We evaluated our method on obtaining Pareto optimal policies using a number of continuous control problems with high degrees of freedom.

The present study aims at investigating how eye contact established by a humanoid robot affects engagement in human-robot interaction (HRI). To this end, we combined explicit subjective evaluations with implicit measures, i.e. reaction times and eye tracking. More specifically, we employed a gaze cueing paradigm in HRI protocol involving the iCub robot. Critically, before moving its gaze, iCub either established eye contact or not with the user. We investigated the patterns of fixations of participants' gaze on the robot's face, joint attention and the subjective ratings of engagement as a function of eye contact or no eye contact. We found that eye contact affected implicit measures of engagement, i.e. longer fixation times on the robot's face during eye contact. Moreover, we showed that joint attention was elicited only when the robot established eye contact, whereas no joint attention occurred when it did not. On the contrary, explicit measures of engagement with the robot did not vary across conditions. Our results highlight the value of combining explicit with implicit measures in an HRI protocol in order to unveil underlying human cognitive mechanisms, which might be at stake during the interactions. These mechanisms could be crucial for establishing an effective and engaging HRI, and provide guidelines to the robotics community with respect to better robot design.

We propose a planning method to enable fast autonomous flight in cluttered environments. Typically, autonomous navigation through a complex environment requires a continuous search on a graph generated by a k-connected grid or a probabilistic scheme. As the vehicle travels, updating the graph with data from onboard sensors is expensive as is the search on the graph especially if the paths must be kinodynamically feasible. We propose to avoid the online search to reduce the computational complexity. Our method models the environment differently in two separate regions. Obstacles are considered to be deterministically known within the sensor range and probabilistically known beyond the sensor range. Instead of searching for the path with the lowest cost (typically the shortest path), the method maximizes the likelihood to reach the goal in determining the immediate next step for navigation. With such a problem formulation, the online method realized by a trajectory library can determine a path within 0.2-0.3ms using a single CPU thread on a modem embedded computer. In experiments, it enables a lightweight UAV to fly at 10m/s in a cluttered forest environment (see Fig. 1 as an example).

This paper considers the problem of planning trajectories for robots equipped with sensors whose task is to track an evolving target process in the world. We focus on processes which can be represented by a Gaussian random variable, which is known to reduce the general stochastic information acquisition problem to a deterministic problem, which is much simpler to solve. Previous work on solving the resulting deterministic problem focuses on computing a search tree by Forward Value Iteration and pruning uninformative nodes early on in the search via a domination criteria. In this work we formulate the Active Information Acquisition problem as a deterministic planning problem where algorithms like Dijkstra and A* can produce optimal solutions. To use A* effectively in long planning horizons we derive a consistent and admissible heuristic as a function of the sensor model which can be used in information acquisition tasks such as actively mapping static and moving targets in an environment with obstacles. We validate the results in several simulations indicating that the resulting heuristic informed algorithm can recover optimal solutions faster than existing search-based methods.

In an autonomous scientific exploration system, the terrain map generated from mapping process integrates sensing information from multiple aspects and lays the base for decision making processes. With the increasing challenges in planetary exploration, equipping planetary rovers with the principles of terramechanics is becoming more and more common, especially on rough or intricate terrain. However, it is difficult for conventional maps with elevation information only to reflect terrain mechanical properties, which play important roles in terramechanics-based simulation or motion control. This study extracts the dominant parameters in terrain bearing and shearing models, and presents a multi-layered grid map with fundamental geometric and mechanical elements. A corresponding mapping scheme based on dense visual input is designed to reconstruct elevation in the map and predict terrain mechanical parameters of the entire visual field. Experiments are conducted to verify the practicability of the approach proposed in a Mars emulation yard with a rover prototype.

Localization is a key capability for autonomous vehicles. High-Definition maps are a popular method to represent the environment and to enable precise localization. However, the creation is very demanding and it is not always guaranteed to receive accurate map information, especially for unstructured areas. In this paper, we introduce a novel probabilistic localization and mapping framework that brings together the advantages of sparse feature maps, multi-target tracking for landmark detection, probabilistic global vehicle localization and a graph-based formulation to achieve a consistent map. The front-end of our Simultaneous Localization and Mapping framework is based on Monte Carlo Localization. Our novel measurement model integrates a virtual topological Path-Map with sparse map features to obtain global localization. The graph-based back-end optimizes online the vehicle trajectory and the landmarks' configuration to create a globally aligned map. Furthermore, our method allows weaker requirements in terms of accuracy of the sparse feature map as we represent the degree of uncertainty by means of probabilistic distribution. Additionally, the sparse feature map representation needs substantially less memory than other approaches, which is an advantage for autonomous vehicles. The framework has been tested and evaluated in real experiments for several autonomous runs. The results demonstrate the robustness of our system.

Mobile service robots deployed in populated environments like train stations, airports or offices are not only required to move safely, but also in socially acceptable ways. In order to achieve this, robots need to be able to track people within their vicinity. This work presents an approach to tracking people from a mobile robot platform, incorporating a novel approach to human motion prediction. Unfavorable viewing angles, motion blur and ever-changing light conditions are constant issues for sensors on mobile vehicles. Therefore, a system is needed which increases the tracking quality of humans in order to cope with a low detection rate. The scientific contribution of this paper lies in a precise human model for tracking which utilizes historical spatial data of pedestrians from previous detection and from simulation. The model is embedded into a particle-filter based tracking approach designed for the use on a moving platform and is able to incorporate a variety of person detectors. Experiments conducted prove that the proposed method increases tracking quality, especially at a low detection rate.

We designed a manipulation purpose small agent vehicle and performed a ghost net recovery mission with the vehicle to test its control accuracy and capacity to be used in other tasks. We constructed a control model that allows the vehicle to perform the complex motion of 4-degrees-of-freedom (DOF) movements and underwater manipulation. Also, we implemented a ghost net recovery algorithm that can automatically detect, grip, and lift ghost nets. The test results show stable yaw, depth, and position control capabilities, and the manipulation performance was verified by automatically performing a ghost net recovery mission.

This paper introduces a taxonomy of manipulations as seen especially in cooking for 1) grouping manipulations from the robotics point of view, 2) consolidating aliases and removing ambiguity for motion types, and 3) provide a path to transferring learned manipulations to new unlearned manipulations. Using instructional videos as a reference, we selected a list of common manipulation motions seen in cooking activities grouped into similar motions based on several trajectory and contact attributes. Manipulation codes are then developed based on the taxonomy attributes to represent the manipulation motions. The manipulation taxonomy is then used for comparing motion data in the Daily Interactive Manipulation (DIM) data set to reveal their motion similarities.

In this paper, we focus on the challenging perception problem in robotic pouring. Most of the existing approaches either leverage visual or haptic information. However, these techniques may suffer from poor generalization performances on opaque containers or concerning measuring precision. To tackle these drawbacks, we propose to make use of audio vibration sensing and design a deep neural network PouringNet to predict the liquid height from the audio fragment during the robotic pouring task. PouringNet is trained on our collected real-world pouring dataset with multimodal sensing data, which contains more than 3000 recordings of audio, force feedback, video and trajectory data of the human hand that performs the pouring task. Each record represents a complete pouring procedure. We conduct several evaluations on PouringNet with our dataset and robotic hardware. The results demonstrate that our PouringNet generalizes well across different liquid containers, positions of the audio receiver, initial liquid heights and types of liquid, and facilitates a more robust and accurate audio-based perception for robotic pouring.

Dynamic self-assembly is treated as a promising approach for generating a robotic swarm to perform coordinated tasks, and the assembled pattern can be tuned by regulating the energy input. However, location of a dynamically assembled pattern is hard to be determined, especially under global fields, such as magnetic field. In this paper, we report the formation and manipulation of dynamic self-assembled droplets at the air-liquid interface with the assistance of a magnetic needle. Affected by the locally induced field gradient near the needle, reconfigurable assembled droplets are obtained with higher time-efficiency, and the location of the pattern can be determined. The pattern is reversibly tuned to exhibit expansion and shrinkage by adjusting the height of the needle. Assembled droplets are able to be steered via following the needle in a controlled manner. Moreover, cargo is trapped by exploiting the induced rotational flow around the droplets, and it can also be caged into the central area of the pattern and transported to the desired location. The proposed method opens new prospects of using energy-dissipative pattern as an untethered end-effector for microrobotic manipulation.

Deep Brain Stimulation is used to treat Parkinson's disease and other neurological disorders by implanting an electrode into the brain via a straight-path needle insertion. Enabling course correction and curved trajectories by using a steerable needle has the potential to improve operative outcomes. In this work, a physically motivated dynamic model for an actively steered magnetic-tipped needle is derived. Process and measurement noise covariances and model parameter curvature gain are identified using an expectation-maximization (EM) algorithm. Parameter convergence and accuracy are evaluated, and an RMS position trajectory accuracy of 0. 81 mm is calculated for expected conditions. The EM algorithm converged for expected parameter variations in simulation, which supports the EM implementation's use in identifying the parameters of the model in a physical system.

Minimally invasive surgery is now a well established field in surgery but continuous efforts are made to reduce invasiveness even further. This paper proposes a novel concept of small-diameter multi-arm robot for SinglePort Access Surgery. The concept introduces a combination of backbone and actuation principles in a macro-micro fashion to achieve an excellent decoupling of the triangulation platform (macro) and of the end-effectors (micro). Concentric tube robots are used for the triangulation platform, while compliant fluidic-actuated bending segments are used as end-effectors. The fluidic actuation is advantageous as it minimally interferes with the triangulation platform. The triangulation platform on the other hand provides a stable base for the end-effectors such that large distal actuation bandwidth can be achieved. A specific embodiment for Spina Bifida repair is developed and proposed. The surgical and technical requirements as well as the mechanical design are presented in details. A first prototype is built and characterization experiments are conducted to evaluate its performance.

Rapidly Random exploring Trees are popular algorithms in the field of motion planning. A feasible path connecting two different poses is found by incrementally building a tree data structure. They are powerful and flexible, but also computationally intense, requiring thousands of iterations before their termination. The aim of this article is to show the capabilities of MT-RRT, a general purpose library which exploits four different multithreading strategies to speed up the planning process of Rapidly Random exploring Trees. MT-RRT will be proved to significantly reduce the computation time on various benchmarks.

Wearable robots suffer from issues of usability, cost-performance, and battery life. These drawbacks hinder the development of their market and their implementation in healthcare and industry. Back support exoskeletons in particular rely heavily on mechanical storing of energy but are still mostly using traditional mechanical elements and actuators to achieve the task. The core technology of our work, named MRLink, uses a smart fluid in combination with a compression spring to produce a unique energy store-and-release functionality. With only 5 Watts MRLink can produce over 500N of braking force. The braking force, variable viscosity, can be continuously controlled through the supply current. With appropriate selection of the spring and control method, the required assist energy can be acquired from body weight and movement dynamics, stored momentarily and, released at the appropriate timing to assist the body motion. The MRLink does not require an external power source like pneumatic actuators, does not require gearboxes and large batteries like DC motors, and comes in a compact package. Thus, this technology has the potential to significantly contribute to wearable exoskeletons in general, and back support exoskeletons in particular. In this work we present the development of a back support exoskeleton prototype based on MRLink, we describe its functional details, and a pilot test partially verifying the proposed functions.

Building small custom robots is getting democratized thanks to affordable tools like 3D printers and micro controllers. However, it still requires expertise from a wide range of domains: from designing the mechanical parts to writing the code that controls the robot. We present MPERL, a tool to help non-experts build custom robotic manipulators. MPERL starts from an abstract description of the robot's kinematic structure which contains information about joints, actuators, and sensors. The structure is refined with an easily manufactured geometry. Furthermore, from the structure MPERL generates control code which can be used to move the robot to a target configuration. We evaluate MPERL on a range of common robotic manipulator architectures, both serial and parallel.