ICRA 2020 论文列表

Taking selfies has become a photographic trend nowadays. We envision the emergence of the "video selfie" capturing a short continuous video clip (or burst photography) of the user, themselves. A selfie stick is usually used, whereby a camera is mounted on a stick for taking selfie photos. In this scenario, we observe that the camera typically goes through a special trajectory along a sphere surface. Motivated by this observation, in this work, we propose an efficient and globally optimal relative camera pose estimation between a pair of two images captured by a camera mounted on a selfie stick. We exploit the special geometric structure of the camera motion constrained by a selfie stick and define its motion as spherical joint motion. By the new parametrization and calibration scheme, we show that the pose estimation problem can be reduced to a 3-DoF (degrees of freedom) search problem, instead of a generic 6-DoF problem. This allows us to derive a fast branch-and-bound global optimization, which guarantees a global optimum. Thereby, we achieve efficient and robust estimation even in the presence of outliers. By experiments on both synthetic and real-world data, we validate the performance as well as the guaranteed optimality of the proposed method.

This paper addresses the generation of collision-free trajectories for the autonomous execution of assistive tasks in Robotic Minimally Invasive Surgery (R-MIS). The proposed approach takes into account geometric constraints related to the desired task, like for example the direction to approach the final target and the presence of moving obstacles. The developed motion planner is structured as a two-layer architecture: a global level computes smooth spline-based trajectories that are continuously updated using virtual potential fields; a local level, exploiting Dynamical Systems based obstacle avoidance, ensures collision free connections among the spline control points. The proposed architecture is validated in a realistic surgical scenario.

Global localization is an important and widely studied problem for many robotic applications. Place recognition approaches can be exploited to solve this task, e.g., in the autonomous driving field. While most vision-based approaches match an image w.r.t. an image database, global visual localization within LiDAR-maps remains fairly unexplored, even though the path toward high definition 3D maps, produced mainly from LiDARs, is clear. In this work we leverage Deep Neural Network (DNN) approaches to create a shared embedding space between images and LiDAR-maps, allowing for image to 3D-LiDAR place recognition. We trained a 2D and a 3D DNN that create embeddings, respectively from images and from point clouds, that are close to each other whether they refer to the same place. An extensive experimental activity is presented to assess the effectiveness of the approach w.r.t. different learning paradigms, network architectures, and loss functions. All the evaluations have been performed using the Oxford Robotcar Dataset, which encompasses a wide range of weather and light conditions.

Traditional robotic control suits require profound task-specific knowledge for designing, building and testing control software. The rise of Deep Learning has enabled end-to-end solutions to be learned entirely from data, requiring minimal knowledge about the application area. We design a learning scheme to train end-to-end linear dynamical systems (LDS)s by gradient descent in imitation learning robotic domains. We introduce a new regularization loss component together with a learning algorithm that improves the stability of the learned autonomous system, by forcing the eigenvalues of the internal state updates of an LDS to be negative reals. We evaluate our approach on a series of real-life and simulated robotic experiments, in comparison to linear and nonlinear Recurrent Neural Network (RNN) architectures. Our results show that our stabilizing method significantly improves test performance of LDS, enabling such linear models to match the performance of contemporary nonlinear RNN architectures. A video of the obstacle avoidance performance of our method on a mobile robot, in unseen environments, compared to other methods can be viewed at https://youtu.be/mhEsCoNao5E.

In many robot control problems, factors such as stiffness and damping matrices and manipulability ellipsoids are naturally represented as symmetric positive definite (SPD) matrices, which capture the specific geometric characteristics of those factors. Typical learned skill models such as dynamic movement primitives (DMPs) can not, however, be directly employed with quantities expressed as SPD matrices as they are limited to data in Euclidean space. In this paper, we propose a novel and mathematically principled framework that uses Riemannian metrics to reformulate DMPs such that the resulting formulation can operate with SPD data in the SPD manifold. Evaluation of the approach demonstrates that beneficial properties of DMPs such as change of the goal during operation apply also to the proposed formulation.

ImageNet-pretrained networks have been widely used in transfer learning for monocular depth estimation. These pretrained networks are trained with classification losses for which only semantic information is exploited while spatial information is ignored. However, both semantic and spatial information is important for per-pixel depth estimation. In this paper, we design a novel self-supervised geometric pretraining task that is tailored for monocular depth estimation using uncalibrated videos. The designed task decouples the structure information from input videos by a simple yet effective conditional autoencoder-decoder structure. Using almost unlimited videos from the internet, networks are pretrained to capture a variety of structures of the scene and can be easily transferred to depth estimation tasks using calibrated images. Extensive experiments are used to demonstrate that the proposed geometric-pretrained networks perform better than ImageNet-pretrained networks in terms of accuracy, few-shot learning and generalization ability. Using existing learning methods, geometric-transferred networks achieve new state-of-the-art results by a large margin. The pretrained networks will be open source soon1 .

This paper considers basket grasps, where a two-finger robot hand forms a basket that can safely lift and carry rigid objects in a 2-D gravitational environment. The two-finger basket grasps form special points in a high-dimensional configuration space of the object and two-finger robot hand. This paper establishes that all two-finger basket grasps can be found in a low-dimensional contact space that parametrizes the two-finger contacts along the supported object boundary. Using contact space, each basket grasp is associated with its depth that provides a security measure while carrying the object, as well as its safety margin away from a critical finger opening where the object drops-off into its intended destination. Geometric techniques that compute the depth and drop-off finger opening are described and illustrated with detailed graphical and numerical examples.

In order to understand how a robot will perform in the open world, we aim to establish a quantitative understanding of the distribution of environments that a robot will face when when it is deployed. However, even restricting attention only to the distribution of objects in a scene, these distributions over environments are nontrivial: they describe mixtures of discrete and continuous variables related to the number, type, poses, and attributes of objects in the scene. We describe a probabilistic generative model that uses scene trees to capture hierarchical relationships between collections of objects, as well as a variational inference algorithm for tuning that model to best match a set of observed environments without any need for tediously labeled parse trees. We demonstrate that this model can accurately capture the distribution of a pair of nontrivial manipulation-relevant datasets and be deployed as a density estimator and outlier detector for novel environments.

In this paper, we consider the generation of generic object candidates by a mobile robot that is endowed with a pan-tilt monocular camera. This is an important problem because these candidates serve as basis for the robot to categorize and/or recognize the objects in its surroundings. The previously proposed methods either do not have a means of enabling the robot to look around through moving its camera or do not take advantage of the temporal coherence of the video data. We present a novel approach that enables the robot to achieve both of these capabilities simultaneously. In this approach, the robot's camera movements are governed by a family of controllers whose constructions depend on the set of object candidates that have been hitherto generated, but not directly looked at. In parallel, the robot discovers the object candidates through tracking segments and determining spatio-temporally coherent ones. The advantage of the proposed approach is that while the robot can explore its surroundings by simply looking around prior to more sophisticated exploration behavior involving possibly bodily locomotion the generated object candidates turn out to be consolidated across the visual stream in comparison to single-shot methods. This is demonstrated in extensive experimental results with a robot operating indoors varying in clutter as well as outdoors.

The safety of an autonomous vehicle not only depends on its own perception of the world around it, but also on the perception and recognition from other vehicles. If an ego vehicle considers the uncertainty other vehicles have about itself, then by reducing the estimated uncertainty it can increase its safety. In this paper, we focus on how an ego vehicle plans its trajectories through the blind spots of other vehicles. We create visibility-aware planning, where the ego vehicle chooses its trajectories such that it reduces the perceived uncertainty other vehicles may have about the state of the ego vehicle. We present simulations of traffic and highway environments, where an ego vehicle must pass another vehicle, make a lane change, or traverse a partially-occluded intersection. Emergent behavior shows that when using visibility-aware planning, the ego vehicle spends less time in a blind spot, and may slow down before entering the blind spot so as to increase the likelihood other vehicles perceive the ego vehicle.

Travel over sloped terrain is difficult as an incline changes the interaction between each wheel and the ground resulting in an unbalanced load distribution which can lead to loss of traction and instability. This paper presents a novel approach to generating wheel rotation for primary locomotion by only changing its centre of rotation, or as a complimentary locomotion source to increase versatility of a plain centre hub drive. This is done using linear actuators within a wheel to control the position of the centre hub and induce a moment on the wheel from gravity. In doing so our platform allows for active ride height selection and individual wheel pose control. We present the system with calculations outlining the theoretical properties and perform experiments to validate the concept under loading via multiple gaits to show motion on slopes, and sustained motion over extended distance. We envision applications in conjunction to assist current motor drives and increasing slope traversability by allowing body pose and centre of gravity manipulation, or as a primary locomotion system.

We propose a generative model for the spatio-temporal distribution of high dimensional categorical observations. These are commonly produced by robots equipped with an imaging sensor such as a camera, paired with an image classifier, potentially producing observations over thousands of categories. The proposed approach combines the use of Dirichlet distributions to model sparse co-occurrence relations between the observed categories using a latent variable, and Gaussian processes to model the latent variable's spatio-temporal distribution. Experiments in this paper show that the resulting model is able to efficiently and accurately approximate the temporal distribution of high dimensional categorical measurements such as taxonomic observations of microscopic organisms in the ocean, even in unobserved (held out) locations, far from other samples. This work's primary motivation is to enable deployment of informative path planning techniques over high dimensional categorical fields, which until now have been limited to scalar or low dimensional vector observations.

This paper presents an approach for implementing game theoretic decision making in combination with realistic sensory data input so as to allow an autonomous vehicle to perform maneuvers, such as lane change or merge in high traffic scenarios. The main novelty of this work, is the use of realistic sensory data input to obtain the observations as input of an iterative multi-player game in a realistic simulator. The game model allows to anticipate reactions of additional vehicles to the movements of the ego-vehicle without using any specific coordination or vehicle-to-vehicle communication. Moreover, direct information from the simulator, such as position or speed of the vehicles is also avoided.The solution of the game is based on cognitive hierarchy reasoning and it uses Monte Carlo reinforcement learning in order to obtain a near-optimal policy towards a specific goal. Moreover, the game proposed is capable of solving different situations using a single policy. The system has been successfully tested and compared with previous techniques using a realistic hybrid simulator, where the ego-vehicle and its sensors are simulated on a 3D simulator and the additional vehicles' behavior is obtained from a traffic simulator.

Since the uprising of new biomedical orthotic devices, exoskeletons have been put in the spotlight for their possible use in rehabilitation. Even if these products might share some commonalities among them in terms of overall structure, degrees of freedom and possible actions, they quite often differ in their approach on how to generate a feasible, stable and comfortable gait trajectory pattern. This paper introduces three proposed trajectories that were generated by using a basis function interpolation method and by working closely with two major rehabilitation centers in Italy. The whole procedure has been focused on the concepts of a configurable walk for patients that suffer from spinal cord injuries. We tested the solutions on a group of healthy volunteers and on a spinal-cord injury patient with the use of the new TWIN exoskeleton developed at the Rehab Technologies Lab at the Italian Institute of Technology.

Ground Penetrating Radar (GPR) is one of the most important non-destructive evaluation (NDE) devices to detect the subsurface objects (i.e. rebars, utility pipes) and reveal the underground scene. One of the biggest challenges in GPR based inspection is the subsurface targets reconstruction. In order to address this issue, this paper presents a 3D GPR migration and dielectric prediction system to detect and reconstruct underground targets. This system is composed of three modules: 1) visual inertial fusion (VIF) module to generate the pose information of GPR device, 2) deep neural network module (i.e., DepthNet) which detects B-scan of GPR image, extracts hyperbola features to remove the noise in B-scan data and predicts dielectric to determine the depth of the objects, 3) 3D GPR migration module which synchronizes the pose information with GPR scan data processed by DepthNet to reconstruct and visualize the 3D underground targets. Our proposed DepthNet processes the GPR data by removing the noise in B-scan image as well as predicting depth of subsurface objects. For DepthNet model training and testing, we collect the real GPR data in the concrete test pit at Geophysical Survey System Inc. (GSSI) and create the synthetic GPR data by using gprMax3.0 simulator. The dataset we create includes 350 labeled GPR images. The DepthNet achieves an average accuracy of 92.64% for B-scan feature detection and an 0.112 average error for underground target depth prediction. In addition, the experimental results verify that our proposed method improve the migration accuracy and performance in generating 3D GPR image compared with the traditional migration methods.

Initialization is essential to monocular Simultaneous Localization and Mapping (SLAM) problems. This paper focuses on a novel initialization method for monocular SLAM based on planar features. The algorithm starts by homography estimation in a sliding window. It then proceeds to a global plane optimization (GPO) to obtain camera poses and the plane normal. 3D points can be recovered using planar constraints without triangulation. The proposed method fully exploits the plane information from multiple frames and avoids the ambiguities in homography decomposition. We validate our algorithm on the collected chessboard dataset against baseline implementations and present extensive analysis. Experimental results show that our method outperforms the ne-tuned baselines in both accuracy and real-time.

Rapid and reliable robot bin picking is a critical challenge in automating warehouses, often measured in picks-per-hour (PPH). We explore increasing PPH using faster motions based on optimizing over a set of candidate grasps. The source of this set of grasps is two-fold: (1) grasp-analysis tools such as Dex-Net generate multiple candidate grasps, and (2) each of these grasps has a degree of freedom about which a robot gripper can rotate. In this paper, we present Grasp-Optimized Motion Planning (GOMP), an algorithm that speeds up the execution of a bin-picking robot's operations by incorporating robot dynamics and a set of candidate grasps produced by a grasp planner into an optimizing motion planner. We compute motions by optimizing with sequential quadratic programming (SQP) and iteratively updating trust regions to account for the non-convex nature of the problem. In our formulation, we constrain the motion to remain within the mechanical limits of the robot while avoiding obstacles. We further convert the problem to a time-minimization by repeatedly shorting a time horizon of a trajectory until the SQP is infeasible. In experiments with a UR5, GOMP achieves a speedup of 9x over a baseline planner.

Path planning algorithms for steerable catheters, must guarantee anatomical obstacles avoidance, reduce the insertion length and ensure the compliance with needle kinematics. The majority of the solutions in literature focuses on graph based or sampling based methods, both limited by the impossibility to directly obtain smooth trajectories. In this work we formulate the path planning problem as a reinforcement learning problem and show that the trajectory planning model, generated from the training, can provide the user with optimal trajectories in terms of obstacle clearance and kinematic constraints. We obtain 2D and 3D environments from MRI images processing and we implement a GA3C algorithm to create a path planning model, able to generalize on different patients anatomies. The curvilinear trajectories obtained from the model in 2D and 3D environments are compared to the ones obtained by A* and RRT* algorithms. Our method achieves state-of-the-art performances in terms of obstacle avoidance, trajectory smoothness and computational time proving this algorithm as valid planning method for complex environments.

Deployment and operation of autonomous underwater vehicles is expensive and time-consuming. High-quality realistic sonar data simulation could be of benefit to multiple applications, including training of human operators for post-mission analysis, as well as tuning and validation of autonomous target recognition (ATR) systems for underwater vehicles. Producing realistic synthetic sonar imagery is a challenging problem as the model has to account for specific artefacts of real acoustic sensors, vehicle attitude, and a variety of environmental factors. We propose a novel method for generating realistic-looking sonar side-scans of full-length missions, called Markov Conditional pix2pix (MC-pix2pix). Quantitative assessment results confirm that the quality of the produced data is almost indistinguishable from real. Furthermore, we show that bootstrapping ATR systems with MC-pix2pix data can improve the performance. Synthetic data is generated 18 times faster than real acquisition speed, with full user control over the topography of the generated data.