IROS 2019 Papers

Service robots are expected to operate effectively in human-centric environments for long periods of time. In such realistic scenarios, fine-grained object categorization is as important as basic-level object categorization. We tackle this problem by proposing an open-ended object recognition approach which concurrently learns both the object categories and the local features for encoding objects. In this work, each object is represented using a set of general latent visual topics and category-specific dictionaries. The general topics encode the common patterns of all categories, while the category-specific dictionary describes the content of each category in details. The proposed approach discovers both sets of general and specific representations in an unsupervised fashion and updates them incrementally using new object views. Experimental results show that our approach yields significant improvements over the previous state-of-the-art approaches concerning scalability and object classification performance. Moreover, our approach demonstrates the capability of learning from very few training examples in a real-world setting. Regarding computation time, the best result was obtained with a Bag-of-Words method followed by a variant of the Latent Dirichlet Allocation approach.

In order for robots to share their workspace with people, they need to reason about human motion efficiently. In this work we leverage large datasets of paths in order to infer local models that are able to perform long-term predictions of human motion. Further, since our method is based on simple dynamics, it is conceptually simple to understand and allows one to interpret the predictions produced, as well as to extract a cost function that can be used for planning. The main difference between our method and similar systems, is that we employ a map of the space and translate the motion of groups of paths into vector fields on that map. We test our method on synthetic data and show its performance on the Edinburgh forum pedestrian long-term tracking dataset [1] where we were able to outperform a Gaussian Mixture Model tasked with extracting dynamics from the paths.

With the advance in the field of mobile robots, autonomous robots are required for long-term deployment in dynamic and complex environments. However, the performance of Visual Inertial SLAM systems in long-term operation is not satisfactory, and most long-term SLAM systems assumes periodic changes in the environment. This paper presents a novel solution for long-term monocular VI SLAM system in dynamic environment based on autoregression(AR) modeling and map prediction. Map points are first classified into static and semi-static map points according to a memory model. Modeling and prediction of the different states of semi-static map points are performed that are derived from time series models. The predicted map is then fused with the current map to achieve a better forecast for the next frame if the prediction is not satisfactory enough. Experiments are carried out on an embedded system. The results indicate that the map prediction is reliable and the proposed approach improves the performance of long-term localization and mapping in dynamic environments.

Learned Neural Network based policies have shown promising results for robot navigation. However, most of these approaches fall short of being used on a real robot due to the extensive simulated training they require. These simulations lack the visuals and dynamics of the real world, which makes it infeasible to deploy on a real robot. We present a novel Neural Net based policy, NavNet, which allows for easy deployment on a real robot. It consists of two sub policies - a high level policy which can understand real images and perform long range planning expressed in high level commands; a low level policy that can translate the long range plan into low level commands on a specific platform in a safe and robust manner. For every new deployment, the high level policy is trained on an easily obtainable scan of the environment modeling its visuals and layout. We detail the design of such an environment and how one can use it for training a final navigation policy. Further, we demonstrate a learned low-level policy. We deploy the model in a large office building and test it extensively, achieving 0.80 success rate over long navigation runs and outperforming SLAM-based models in the same settings.

This paper focuses on building semantic maps, containing object poses and shapes, using a monocular camera. This is an important problem because robots need rich understanding of geometry and context if they are to shape the future of transportation, construction, and agriculture. Our contribution is an instance-specific mesh model of object shape that can be optimized online based on semantic information extracted from camera images. Multi-view constraints on the object shape are obtained by detecting objects and extracting category-specific keypoints and segmentation masks. We show that the errors between projections of the mesh model and the observed keypoints and masks can be differentiated in order to obtain accurate instance-specific object shapes. We evaluate the performance of the proposed approach in simulation and on the KITTI dataset by building maps of car poses and shapes.

In this paper, we propose a novel pose optimizer which can be inserted into either supervised or unsupervised end-to-end visual odometry for the purpose of local pose optimization. The pose optimizer is an analogue of the pose graph optimization used in traditional VSLAM algorithms. Local pose optimization is performed by an attention-based neural network which iteratively refines the predicted pose estimates of an image snippet. Instead of complicated graph convolutional network, the attention mechanism based on geometric consistency of trajectory constraint is utilized because pose features whose spatial distribution is not important can be flattened to vectors and then processed. The pose optimizer is aimed at improving pose estimation accuracy by redistributing errors of pose estimates. Quantitative and qualitative evaluation of the proposed approach on the KITTI Odometry dataset [1] is presented to demonstrate its effectiveness in improving pose estimation accuracy and minimizing pose drift.

Motor babbling and goal babbling has been used for sensorimotor learning of highly redundant systems in soft robotics. Recent works in goal babbling have demonstrated successful learning of inverse kinematics (IK) on such systems, and suggest that babbling in the goal space better resolves motor redundancy by learning as few yet efficient sensorimotor mappings as possible. However, for musculoskeletal robot systems, motor redundancy can provide useful information to explain muscle activation patterns, thus the term motor abundance. In this work, we introduce some simple heuristics to empirically define the unknown goal space, and learn the IK of a 10 DoF musculoskeletal robot arm using directed goal babbling. We then further propose local online motor babbling guided by Covariance Matrix Adaptation Evolution Strategy (CMA-ES), which bootstraps on the goal babbling samples for initialization, such that motor abundance can be queried online for any static goal. Our approach leverages the resolving of redundancies and the efficient guided exploration of motor abundance in two stages of learning, allowing both kinematic accuracy and motor variability at the queried goal. The result shows that local online motor babbling guided by CMA-ES can efficiently explore motor abundance at queried goal positions on a musculoskeletal robot system and gives useful insights in terms of muscle stiffness and synergy.

3D lines in structured environments encode particular regularity like parallelism and orthogonality. We leverage this structural regularity to estimate the absolute and relative camera poses. We decouple the rotation and translation, and propose a novel rotation estimation method. We decompose the absolute and relative rotations and reformulate the problem as computing the rotation from the Manhattan frame to the camera frame. To compute this rotation, we propose an accurate and efficient two-step method. We first estimate its two degrees of freedom (DOF) by two image lines, and then estimate its third DOF by another image line. For these lines, we assume their associated 3D lines are mutually orthogonal, or two 3D lines are parallel to each other and orthogonal to the third. Thanks to our two-step DOF estimation, our absolute and relative pose estimation methods are accurate and efficient. Moreover, our relative pose estimation method relies on weaker assumptions or less correspondences than existing approaches. We also propose a novel strategy to reject outliers and identify dominant directions of the scene. We integrate it into our pose estimation methods, and show that it is more robust than RANSAC. Experiments on synthetic and real-world datasets demonstrated that our methods outperform state-of-the-art approaches.

This paper was motivated by the problem of how to make robots fuse and transfer their experience so that they can effectively use prior knowledge and quickly adapt to new environments. To address the problem, we present a learning architecture for navigation in cloud robotic systems: Lifelong Federated Reinforcement Learning (LFRL). In the work, we propose a knowledge fusion algorithm for upgrading a shared model deployed on the cloud. Then, effective transfer learning methods in LFRL are introduced. LFRL is consistent with human cognitive science and fits well in cloud robotic systems. Experiments show that LFRL greatly improves the efficiency of reinforcement learning for robot navigation. The cloud robotic system deployment also shows that LFRL is capable of fusing prior knowledge. In addition, we release a cloud robotic navigation-learning website to provide the service based on LFRL: www.shared-robotics.com.

We propose a new approach called LiDAR-Flow to robustly estimate a dense scene flow by fusing a sparse LiDAR with stereo images. We take the advantage of the high accuracy of LiDAR to resolve the lack of information in some regions of stereo images due to textureless objects, shadows, ill-conditioned light environment and many more. Additionally, this fusion can overcome the difficulty of matching unstructured 3D points between LiDAR-only scans. Our LiDAR-Flow approach consists of three main steps; each of them exploits LiDAR measurements. First, we build strong seeds from LiDAR to enhance the robustness of matches between stereo images. The imagery part seeks the motion matches and increases the density of scene flow estimation. Then, a consistency check employs LiDAR seeds to remove the possible mismatches. Finally, LiDAR measurements constraint the edge-preserving interpolation method to fill the remaining gaps. In our evaluation we investigate the individual processing steps of our LiDAR-Flow approach and demonstrate the superior performance compared to image-only approach.

Detection of the navigable regions for the unmanned surface vehicles (USVs) sailing on the narrow rivers is very important. Existing detection methods mostly depend on the cameras, which is sensitive to environments and cannot provide reliable navigable regions for sailing. In this paper, we propose a scheme to process 3D LiDAR data to achieve an accurate and robust navigable regions detection. We conduct field experiments in a narrow river in different scenarios to prove the performance of the proposed scheme, which reaches on average 93.8% precision and 92.7% recall.

This paper presents a Learning-based Nonlinear Model Predictive Control (LB-NMPC) algorithm for reconfigurable autonomous vessels to facilitate high-accurate path tracking. Each vessel is designed to latch to a pre-defined point of another vessel that allows the vessels to form a rigid body. The number of possible configurations of such vessels exponentially grows as the total number of vessels increases, which imposes a technical challenge in modeling and identification. In this work, we propose a framework consisting of a real-time parameter estimator and a feedback control strategy, which is capable of ensuring high-accurate path tracking for any feasible configuration of vessels. Novelty of our method is in that the parameter is estimated on-line and adjusts control parameters (e.g., cost function and dynamic model) simultaneously to improve path-tracking performance. Through experiments on different configurations of connected-vessels, we demonstrate stability of our proposed approach and its effectiveness in high-accuracy in path tracking.

Imitation learning offers a general framework where robots can efficiently acquire novel motor skills from demonstrations of a human teacher. While many promising achievements have been shown, the majority of them are only focused on single-stroke movements, without taking into account the problem of multi-tasks sequencing. Conceivably, sequencing different atomic tasks can further augment the robot's capabilities as well as avoid repetitive demonstrations. In this paper, we propose to address the issue of multi-tasks sequencing with emphasis on handling the so-called competing constraints, which emerge due to the existence of the concurrent constraints from Cartesian and joint trajectories. Specifically, we explore the null space of the robot from an information-theoretic perspective in order to maintain imitation fidelity during transition between consecutive tasks. The effectiveness of the proposed method is validated through simulated and real experiments on the iCub humanoid robot.

Upon the advent of Industry 4.0, collaborative robotics and intelligent automation gain more and more traction for enterprises to improve their production processes. In order to adapt to this trend, new programming, learning and collaborative techniques are investigated. Program-bydemonstration is one of the techniques that aim to reduce the burden of manually programming collaborative robots. However, this is often limited to teaching to grasp at a certain position, rather than grasping a certain object. In this paper, we propose a method that learns to grasp an arbitrary object from visual input. While other learning-based approaches for robotic grasping require collecting a large dataset, manually or automatically labeled in a real or simulated world, our approach requires a single demonstration. We present results on grasping various objects with the Franka Panda collaborative robot after capturing a single image from a wrist mounted RGB camera. From this image we learn a robot controller with a convolutional neural network to adapt to changes in the object's position and rotation with less than 5 minutes of training time on a NVIDIA Titan X GPU, achieving over 90% grasp success rate.

Referring to objects in a natural and unambiguous manner is crucial for effective human-robot interaction. Previous research on learning-based referring expressions has focused primarily on comprehension tasks, while generating referring expressions is still mostly limited to rule-based methods. In this work, we propose a two-stage approach that relies on deep learning for estimating spatial relations to describe an object naturally and unambiguously with a referring expression. We compare our method to the state of the art algorithm in ambiguous environments (e.g., environments that include very similar objects with similar relationships). We show that our method generates referring expressions that people find to be more accurate (~30% better) and would prefer to use (~32% more often).

Model free reinforcement learning suffers from the high sampling complexity inherent to robotic manipulation or locomotion tasks. Most successful approaches typically use random sampling strategies which leads to slow policy convergence. In this paper we present a novel approach for efficient exploration that leverages previously learned tasks. We exploit the fact that the same system is used across many tasks and build a generative model for exploration based on data from previously solved tasks to improve learning new tasks. The approach also enables continuous learning of improved exploration strategies as novel tasks are learned. Extensive simulations on a robot manipulator performing a variety of motion and contact interaction tasks demonstrate the capabilities of the approach. In particular, our experiments suggest that the exploration strategy can more than double learning speed, especially when rewards are sparse. Moreover, the algorithm is robust to task variations and parameter tuning, making it beneficial for complex robotic problems.

We develop a system for modeling hand-object interactions in 3D from RGB images that show a hand which is holding a novel object from a known category. We design a Convolutional Neural Network (CNN) for Hand-held Object Pose and Shape estimation called HOPS-Net and utilize prior work to estimate the hand pose and configuration. We leverage the insight that information about the hand facilitates object pose and shape estimation by incorporating the hand into both training and inference of the object pose and shape as well as the refinement of the estimated pose. The network is trained on a large synthetic dataset of objects in interaction with a human hand. To bridge the gap between real and synthetic images, we employ an image-to-image translation model (Augmented CycleGAN) that generates realistically textured objects given a synthetic rendering. This provides a scalable way of generating annotated data for training HOPS-Net. Our quantitative experiments show that even noisy hand parameters significantly help object pose and shape estimation. The qualitative experiments show results of pose and shape estimation of objects held by a hand “in the wild”.

This paper introduces a reinforcement learning algorithm with robot manipulation to learn an arbitrary object's center of mass whose physical material composition is unknown. Robot learning is through manipulation of the object in a sequence of actions. The effectiveness of the algorithm is demonstrated in simulation to locate the centers of mass of rocks with complex shapes, with even or uneven mass distributions, and confirmed by vertically stacking the rocks along their learned centers of mass both in simulation and in real experiments.

Vision and learning have made significant progress that could improve robotics policies for complex tasks and environments. Learning deep neural networks for image understanding, however, requires large amounts of domain-specific visual data. While collecting such data from real robots is possible, such an approach limits the scalability as learning policies typically requires thousands of trials. In this work we attempt to learn manipulation policies in simulated environments. Simulators enable scalability and provide access to the underlying world state during training. Policies learned in simulators, however, do not transfer well to real scenes given the domain gap between real and synthetic data. We follow recent work on domain randomization and augment synthetic images with sequences of random transformations. Our main contribution is to optimize the augmentation strategy for sim2real transfer and to enable domain-independent policy learning. We design an efficient search for depth image augmentations using object localization as a proxy task. Given the resulting sequence of random transformations, we use it to augment synthetic depth images during policy learning. Our augmentation strategy is policy-independent and enables policy learning with no real images. We demonstrate our approach to significantly improve accuracy on three manipulation tasks evaluated on a real robot.