IROS 2019 论文列表

Precise knowledge of pose is of great importance for reliable operation of mobile robots in outdoor environments. Simultaneous localization and mapping (SLAM) is the online construction of a map during exploration of an environment. One of the components of SLAM is loop closure detection, identifying that the same location has been visited and is present on the existing map, and localizing against it. We have shown in previous work that using semantics from a deep segmentation network in conjunction with the Normal Distributions Transform (NDT) point cloud registration improves the robustness, speed and accuracy of lidar odometry. In this work we extend the method for loop closure detection, using the labels already available from local registration into NDT Histograms, and we present a SLAM pipeline based on Semantic assisted NDT and PointNet ++. We experimentally demonstrate on sequences from the KITTI benchmark that the map descriptor we propose outperforms NDT Histograms without semantics, and we validate its use on a SLAM task.

In this paper, we enhance our knowledge-based and constraint-based approach of robot programming with the concept of Semantic Mates. They describe intended mechanical connections between parts of an assembly. This allows deriving appropriate assembly poses from the type of connection and the geometric properties of the involved parts. The paper presents an ontology-based representation of Semantic Mates that is used to augment object models with additional information regarding their potential use in an assembly. Such semantically annotated object models can be used in our instruction framework to program a robot to perform assembly tasks through simple drag-and-drop operations in a graphical user interface. We conducted a user study with 21 participants in order to evaluate the efficiency and usability of the Semantic Mates concept based on a use-case from the domain of mechanical assembly. Across different experience levels in robotics, the participants achieved a significantly faster workflow and improved perceived usability compared to the manual specification of constraint-based assembly operations.

Instance segmentation of unknown objects from images is regarded as relevant for several robot skills including grasping, tracking and object sorting. Recent results from computer vision have shown that large hand-labeled datasets enable high segmentation performance. To overcome the time-consuming process of manually labeling data for new environments, we present a transfer learning approach for robots that learn to segment objects by interacting with their environment in a self-supervised manner. Our robot pushes unknown objects on a table and uses information from optical flow to create training labels given by object masks. To achieve this, we fine-tune an existing DeepMask instance segmentation network on the self-labeled training data acquired by the robot. We evaluate our trained network (SelfDeepMask) on a set of real images showing challenging and cluttered scenes with novel objects. Here, SelfDeepMask outperforms the DeepMask network trained on the COCO dataset by 8.6% in average precision.

We present a convolutional neural network for joint 3D shape prediction and viewpoint estimation from a single input image. During training, our network gets the learning signal from a silhouette of an object in the input image-a form of self-supervision. It does not require ground truth data for 3D shapes and the viewpoints. Because it relies on such a weak form of supervision, our approach can easily be applied to real-world data. We demonstrate that our method produces reasonable qualitative and quantitative results on natural images for both shape estimation and viewpoint prediction. Unlike previous approaches, our method does not require multiple views of the same object instance in the dataset, which significantly expands the applicability in practical robotics scenarios. We showcase it by using the hallucinated shapes to improve the performance on the task of grasping real-world objects both in simulation and with a PR2 robot.

In this paper, a self-modeling method based on a causal network is proposed for the tracking control of the Crawler Fire Fighting Robot (CFFR). The method mainly consists of two parts, one is a motion model, based on data driving, learning to establish the correspondence between control signal sequence and vehicle motion, estimating the motion state of the next moment from historical data, eliminating complex CFFR modeling. The other is the tracking network. Based on the simulation data of the motion model, the relationship between the target trajectory and the current control command is learned, which simplifies the design and cumbersome tuning of the complex controller. The effectiveness of the proposed method is verified in both simulated and real-world environments. Qualitative and quantitative experimental results verify the accuracy of the tracking.

Using multi-robot systems for autonomous construction allows for parallelization and scalability. Swarm construction furthermore exploits robot interactions and collaboration, such that the robot swarm collectively constructs artifacts beyond what a single comparable robot could achieve. Here we present an alternative concept of swarm construction that is distinct because it uses continuous building material. Our approach is unique in its use of braiding techniques for construction. We deploy fibres that potentially allow for structures that are not possible with building blocks. To achieve maximal scalability we restrict ourselves to a decentralized approach. The main challenges are the local coordination of the robot teams, self-organized task allocation, and the dynamic reconfiguration of the braiding scheme at runtime. We successfully validate our approach in multi-robot experiments that show both braiding and branching of the braid. In addition, we show options for implementing an open system-that is robots can join and leave the braiding process on the fly.

A lot of research has been conducted on swarm and modular robots and their reconfiguration. For this kind of robot system, it is best to simplify each individual module in order to avoid making the swarm system too complex in the interest of reducing the computing power necessary for a task.This paper proposes a swarm robot system capable of self-segregation by changing the size of each agent, utilising the Brazil Nut Effect. The self-segregation allowed by this effect lead to the simplification of the control protocol necessary for the swarm.We simulated a swarm robot system consisting of circular modules capable of changing their radius and, with the aid of externally supplied vibrations, utilising the Brazil Nut Effect to achieve segregation of the swarm. The effect also allowed for the migration of individual modules through the swarm. Additionally, we harnessed this segregation effect in simulation to create a flow of modules within the swarm, whilst the swarm was confined within a vibrating container.Lastly, we took the first steps toward physical verification of the experimental results by designing and building a prototype for a circular robot module capable of changing its radius in a manner inspired by a spiral torsion spring.

The Spring-Loaded Inverted Pendulum (SLIP) has been widely exploited in both biomechanical and robotics research due to its simple form in mathematics and high accuracy in fitting experimental biology data. However the intrinsic nonlinearity of the SLIP dynamics makes accurate analytical representation unavailable. Traditional methods take advantage of numerical integration to handle this issue while several existing analytical approximations focusing on 2D-SLIP model. The 3D-SLIP suitable to physical reality is rarely investigated. This paper presents a novel perturbation-based approach to obtain the closed-form analytical approximations of the 3D-SLIP model in stance phase. In contrast to existing work ignoring the gravitational forces, the proposed approach just relies on assumptions of small leg compression and small leg swept angle. The performance of the derived approximations has been evaluated via comprehensive numerical analysis. The quality of accurate apex prediction promises the approximation as an advantageous and reliable tool for locomotion control of legged robots.

While convolutional neural networks are dominating the field of computer vision, one usually does not have access to the large amount of domain-relevant data needed for their training. Therefore, it has become common practice to use available synthetic samples along domain adaptation schemes to prepare algorithms for the target domain. Tackling this problem from a different angle, we introduce a pipeline to map unseen target samples into the synthetic domain used to train task-specific methods. Denoising the data and retaining only the features these recognition algorithms are familiar with, our solution greatly improves their performance. As this mapping is easier to learn than the opposite one (i.e., to generate realistic features to augment the source samples), we demonstrate how our whole solution can be trained purely on augmented synthetic data and still performs better than methods trained with domain-relevant information (e.g., real images or realistic textures for the 3D models). Applying our approach to object recognition from texture-less CAD data, we present a custom generative network which fully utilizes the purely geometrical information to learn robust features and to achieve a more refined mapping for unseen color images.

Semantic segmentation and instance level segmentation made substantial progress in recent years due to the emergence of deep neural networks (DNNs). A number of deep architectures with Convolution Neural Networks (CNNs) were proposed that surpass the traditional machine learning approaches for segmentation by a large margin. These architectures predict the directly observable semantic category of each pixel by usually optimizing a cross-entropy loss. In this work we push the limit of semantic segmentation towards predicting semantic labels of directly visible as well as occluded objects or objects parts, where the network's input is a single depth image. We group the semantic categories into one background and multiple foreground object groups, and we propose a modification of the standard cross-entropy loss to cope with the settings. In our experiments we demonstrate that a CNN trained by minimizing the proposed loss is able to predict semantic categories for visible and occluded object parts without requiring to increase the network size (compared to a standard segmentation task). The results are validated on a newly generated dataset (augmented from SUNCG) dataset.

Many existing mobile service robots, including the robots in Robocup@Home, perform their designated tasks only when the robots are stationary. The efficiency of these robots can be improved if they can perform some tasks while moving. In this paper, we propose the concept of mobile workstations, which combine mobile platforms with production machinery to increase efficiency by overlapping production time and delivery time. We present a model of mobile workstations and their jobs and describe the task planning algorithm for a team of mobile workstations. The temporal planning problem for mobile workstations combines both features of job shop scheduling problems (JSP) and traveling salesman problem (TSP), but there is little work in the literature that tackles both JSP and TSP simultaneously. Our first algorithm is a complete search algorithm which returns an optimal temporal plan with minimum makespan, and our second algorithm conducts a local search in the space of task graphs so as to quickly return suboptimal temporal plans. According to our experiments, when the number of jobs is small, our second algorithm can generate near-optimal temporal plans, and when the number of jobs is large, our algorithm can generate much shorter plans than SGPlan 5 and a version of job shop scheduling algorithms.

Large, richly annotated datasets have accelerated progress in fields such as computer vision and natural language processing, but replicating these successes in robotics has been challenging. While prior data collection methodologies such as self-supervision have resulted in large datasets, the data can have poor signal-to-noise ratio. By contrast, previous efforts to collect task demonstrations with humans provide better quality data, but they cannot reach the same data magnitude. Furthermore, neither approach places guarantees on the diversity of the data collected, in terms of solution strategies. In this work, we leverage and extend the RoboTurk platform to scale up data collection for robotic manipulation using remote teleoperation. The primary motivation for our platform is two-fold: (1) to address the shortcomings of prior work and increase the total quantity of manipulation data collected through human supervision by an order of magnitude without sacrificing the quality of the data and (2) to collect data on challenging manipulation tasks across several operators and observe a diverse set of emergent behaviors and solutions. We collected over 111 hours of robot manipulation data across 54 users and 3 challenging manipulation tasks in 1 week, resulting in the largest robot dataset collected via remote teleoperation. We evaluate the quality of our platform, the diversity of demonstrations in our dataset, and the utility of our dataset via quantitative and qualitative analysis. For additional results, supplementary videos, and to download our dataset, visit http;//roboturk.stanford.edu/realrobotdataset.

Distributed self-reconfiguration in large-scale modular robots is a slow process and increasing its speed a major challenge. In this article, we propose an improved and asynchronous version of a previously proposed distributed self-reconfiguration algorithm to build a parametric scaffolding structure. This scaffold can then be coated to form the desired final object. The scaffolding is built through a continuous feeding of modules into the growing shape from an underneath reserve of modules which shows a reconfiguration time improved by a factor of 3√N compared to the previous and synchronous version of the algorithm, therefore attaining an O(N1/3) reconfiguration time, with N the number of modules in the system. Our algorithm uses a local motion coordination algorithm and pipelining techniques to ensure that modules can traverse the structure without collisions or creating deadlocks. Last but not least, our algorithm manages uncertainty in the motion duration of modules without negatively impacting reconfiguration time.

Sampling-based motion planners are often used to solve motion planning problems for robots with many degrees of freedom. These planners explore the related configuration space by random sampling. The well-known issue of the sampling-based planners is the narrow passage problem. Narrow passages are small collision-free regions in the configuration space that are, due to their volume, difficult to cover by the random samples. The volume of the narrow passages can be artificially increased by reducing the size of the robot, e.g., by scaling-down its geometry, which increases the probability of placing the random samples into the narrow passages. This allows us to find an approximate solution (trajectory) and use it as a guide to find the solution for a larger robot. Guiding along an approximate solution may, however, fail if this solution leads through such parts of the configuration space that are not reachable or traversable by a larger robot. To improve this guiding process, we propose to compute several approximate solutions leading through different parts of the configuration space, and use all of them to guide the search for a larger robot. We introduce the concept of disabled regions that are prohibited from the exploration using the sampling process. The disabled regions are defined using trajectories already found in the space being searched. The proposed method can solve planning problems with narrow passages with higher success rate than other state-of-the-art planners.

This paper presents a motion planning approach for an aerial pick-and-place task where an aerial manipulator is supposed to pick up or place an object at locations specified as way-points. In particular, we focus on situations where such way-point constraints are imposed on certain partial state variables, rather than on full state variables. Our proposed framework, based on rapidly exploring random trees star (RRT*) in a bidirectional manner, enables an aerial manipulator to find an optimal trajectory that satisfies way-point constraints with only partial specifications. Here, we suggest an extra merging process to integrate the trees, each originated from the start and goal point. In the merging process, we search various candidate points satisfying a given condition that partially constrains state variables, and select a way-point with full specifications optimal in the perspective of the entire trajectory. Simulation and experiment results are included to validate the proposed framework.

Deep reinforcement learning has proven to be a great success in allowing agents to learn complex tasks. However, its application to actual robots can be prohibitively expensive. Furthermore, the unpredictability of human behavior in human-robot interaction tasks can hinder convergence to a good policy. In this paper, we present an architecture that allows agents to learn models of stochastic environments and use them to accelerate learning. We descirbe how an environment model can be learned online and used to generate synthetic transitions, as well as how an agent can leverage these synthetic data to accelerate learning. We validate our approach using an experiment in which a robotic arm has to complete a task composed of a series of actions based on human gestures. Results show that our approach leads to significantly faster learning, requiring much less interaction with the environment. Furthermore, we demonstrate how learned models can be used by a robot to produce optimal plans in real world applications.

The objective of this paper is to develop a sample efficient end-to-end deep learning method for self-driving cars, where we attempt to increase the value of the information extracted from samples, through careful analysis obtained from each call to expert driver's policy. End-to-end imitation learning is a popular method for computing self-driving car policies. The standard approach relies on collecting pairs of inputs (camera images) and outputs (steering angle, etc.) from an expert policy and fitting a deep neural network to this data to learn the driving policy. Although this approach had some successful demonstrations in the past, learning a good policy might require a lot of samples from the expert driver, which might be resource-consuming. In this work, we develop a novel framework based on the Safe Dataset Aggregation (safe DAgger) approach, where the current learned policy is automatically segmented into different trajectory classes, and the algorithm identifies trajectory segments/classes with the weak performance at each step. Once the trajectory segments with weak performance identified, the sampling algorithm focuses on calling the expert policy only on these segments, which improves the convergence rate. The presented simulation results show that the proposed approach can yield significantly better performance compared to the standard Safe DAgger algorithm while using the same amount of samples from the expert.

So-called collaborative robots are a current trend in industrial robotics. However, they still face many problems in practical application such as reduced speed to ascertain their collaborativeness. The standards prescribe two regimes: (i) speed and separation monitoring and (ii) power and force limiting, where the former requires reliable estimation of distances between the robot and human body parts and the latter imposes constraints on the energy absorbed during collisions prior to robot stopping. Following the standards, we deploy the two collaborative regimes in a single application and study the performance in a mock collaborative task under the individual regimes, including transitions between them. Additionally, we compare the performance under “safety zone monitoring” with keypoint pair-wise separation distance assessment relying on an RGB-D sensor and skeleton extraction algorithm to track human body parts in the workspace. Best performance has been achieved in the following setting: robot operates at full speed until a distance threshold between any robot and human body part is crossed; then, reduced robot speed per power and force limiting is triggered. Robot is halted only when the operator's head crosses a predefined distance from selected robot parts. We demonstrate our methodology on a setup combining a KUICA LBR iiwa robot, Intel RealSense RGB-D sensor and OpenPose for human pose estimation.

Government data identifies driver behaviour errors as a factor in 94% of car crashes, and autonomous vehicles (AVs), which avoids risky driver behaviours completely, are expected to reduce the number of road crashes significantly. Thus, one of the central focuses of developing AVs is to ensure safety during navigation. However, in reality, AV safety has been far below its expectation, and so far, no government has allowed for complete autonomous driving without human supervision. This paper proposes a dynamic safe path planning algorithm for AVs with Gaussian process regulated risk map. By reasonably assuming that the output of the object detection and tracking module follows a multi-variate Gaussian distribution, we put forward a safe path planning paradigm with Gaussian process regulated risk map, ensuring safety with high confidence. Both simulation results and in-vehicle tests demonstrate the effectiveness of the proposed algorithm.