Learning Where to Look: Self-supervised Viewpoint Selection for Active Localization using Geometrical Information

Accurate localization in diverse environments is a fundamental challenge in computer vision and robotics. The task involves determining a sensor's precise position and orientation, typically a camera, within a given space. Traditional localization methods often rely on passive sensing, which may struggle in scenarios with limited features or dynamic environments. In response, this paper explores the domain of active localization, emphasizing the importance of viewpoint selection to enhance localization accuracy. Our contributions involve using a data-driven approach with a simple architecture designed for real-time operation, a self-supervised data training method, and the capability to consistently integrate our map into a planning framework tailored for real-world robotics applications. Our results demonstrate that our method performs better than the existing one, targeting similar problems and generalizing on synthetic and real data. We also release an open-source implementation to benefit the community.

MAD-ICP: It Is All About Matching Data -- Robust and Informed LiDAR Odometry

LiDAR odometry is the task of estimating the ego-motion of the sensor from sequential laser scans. This problem has been addressed by the community for more than two decades, and many effective solutions are available nowadays. Most of these systems implicitly rely on assumptions about the operating environment, the sensor used, and motion pattern. When these assumptions are violated, several well-known systems tend to perform poorly. This paper presents a LiDAR odometry system that can overcome these limitations and operate well under different operating conditions while achieving performance comparable with domain-specific methods. Our algorithm follows the well-known ICP paradigm that leverages a PCA-based kd-tree implementation that is used to extract structural information about the clouds being registered and to compute the minimization metric for the alignment. The drift is bound by managing the local map based on the estimated uncertainty of the tracked pose. To benefit the community, we release an open-source C++ anytime real-time implementation.

VBR: A Vision Benchmark in Rome

This paper presents a vision and perception research dataset collected in Rome, featuring RGB data, 3D point clouds, IMU, and GPS data. We introduce a new benchmark targeting visual odometry and SLAM, to advance the research in autonomous robotics and computer vision. This work complements existing datasets by simultaneously addressing several issues, such as environment diversity, motion patterns, and sensor frequency. It uses up-to-date devices and presents effective procedures to accurately calibrate the intrinsic and extrinsic of the sensors while addressing temporal synchronization. During recording, we cover multi-floor buildings, gardens, urban and highway scenarios. Combining handheld and car-based data collections, our setup can simulate any robot (quadrupeds, quadrotors, autonomous vehicles). The dataset includes an accurate 6-dof ground truth based on a novel methodology that refines the RTK-GPS estimate with LiDAR point clouds through Bundle Adjustment. All sequences divided in training and testing are accessible through our website.

Ca$^2$Lib: Simple and Accurate LiDAR-RGB Calibration using Small Common Markers

In many fields of robotics, knowing the relative position and orientation between two sensors is a mandatory precondition to operate with multiple sensing modalities. In this context, the pair LiDAR-RGB cameras offer complementary features: LiDARs yield sparse high quality range measurements, while RGB cameras provide a dense color measurement of the environment. Existing techniques often rely either on complex calibration targets that are expensive to obtain, or extracted virtual correspondences that can hinder the estimate's accuracy. In this paper we address the problem of LiDAR-RGB calibration using typical calibration patterns (i.e. A3 chessboard) with minimal human intervention. Our approach exploits the planarity of the target to find correspondences between the sensors measurements, leading to features that are robust to LiDAR noise. Moreover, we estimate a solution by solving a joint non-linear optimization problem. We validated our approach by carrying on quantitative and comparative experiments with other state-of-the-art approaches. Our results show that our simple schema performs on par or better than other approches using complex calibration targets. Finally, we release an open-source C++ implementation at \url{https://github.com/srrg-sapienza/ca2lib}

How-to Augmented Lagrangian on Factor Graphs

Factor graphs are a very powerful graphical representation, used to model many problems in robotics. They are widely spread in the areas of Simultaneous Localization and Mapping (SLAM), computer vision, and localization. In this paper we describe an approach to fill the gap with other areas, such as optimal control, by presenting an extension of Factor Graph Solvers to constrained optimization. The core idea of our method is to encapsulate the Augmented Lagrangian (AL) method in factors of the graph that can be integrated straightforwardly in existing factor graph solvers. We show the generality of our approach by addressing three applications, arising from different areas: pose estimation, rotation synchronization and Model Predictive Control (MPC) of a pseudo-omnidirectional platform. We implemented our approach using C++ and ROS. Besides the generality of the approach, application results show that we can favorably compare against domain specific approaches.

Photometric LiDAR and RGB-D Bundle Adjustment

The joint optimization of the sensor trajectory and 3D map is a crucial characteristic of Simultaneous Localization and Mapping (SLAM) systems. To achieve this, the gold standard is Bundle Adjustment (BA). Modern 3D LiDARs now retain higher resolutions that enable the creation of point cloud images resembling those taken by conventional cameras. Nevertheless, the typical effective global refinement techniques employed for RGB-D sensors are not widely applied to LiDARs. This paper presents a novel BA photometric strategy that accounts for both RGB-D and LiDAR in the same way. Our work can be used on top of any SLAM/GNSS estimate to improve and refine the initial trajectory. We conducted different experiments using these two depth sensors on public benchmarks. Our results show that our system performs on par or better compared to other state-of-the-art ad-hoc SLAM/BA strategies, free from data association and without making assumptions about the environment. In addition, we present the benefit of jointly using RGB-D and LiDAR within our unified method. We finally release an open-source CUDA/C++ implementation.

On Domain-Specific Pre-Training for Effective Semantic Perception in Agricultural Robotics

Agricultural robots have the prospect to enable more efficient and sustainable agricultural production of food, feed, and fiber. Perception of crops and weeds is a central component of agricultural robots that aim to monitor fields and assess the plants as well as their growth stage in an automatic manner. Semantic perception mostly relies on deep learning using supervised approaches, which require time and qualified workers to label fairly large amounts of data. In this paper, we look into the problem of reducing the amount of labels without compromising the final segmentation performance. For robots operating in the field, pre-training networks in a supervised way is already a popular method to reduce the number of required labeled images. We investigate the possibility of pre-training in a self-supervised fashion using data from the target domain. To better exploit this data, we propose a set of domain-specific augmentation strategies. We evaluate our pre-training on semantic segmentation and leaf instance segmentation, two important tasks in our domain. The experimental results suggest that pre-training with domain-specific data paired with our data augmentation strategy leads to superior performance compared to commonly used pre-trainings. Furthermore, the pre-trained networks obtain similar performance to the fully supervised with less labeled data.

Low Frequency Spinning LiDAR De-Skewing

Most commercially available Light Detection and Ranging (LiDAR)s measure the distances along a 2D section of the environment by sequentially sampling the free range along directions centered at the sensor's origin. When the sensor moves during the acquisition, the measured ranges are affected by a phenomenon known as skewing, which appears as a distortion in the acquired scan. Skewing potentially affects all systems that rely on LiDAR data, however it could be compensated if the position of the sensor were known each time a single range is measured. Most methods to de-skew a LiDAR are based on external sensors such as IMU or wheel odometry, to estimate these intermediate LiDAR positions. In this paper we present a method that relies exclusively on range measurements to effectively estimate the robot velocities which are then used for de-skewing. Our approach is suitable for low frequency LiDAR where the skewing is more evident. It can be seamlessly integrated into existing pipelines, enhancing their performance at negligible computational cost. We validated the proposed method with statistical experiments characterizing different operating conditions

Handling Constrained Optimization in Factor Graphs for Autonomous Navigation

Factor graphs are graphical models used to represent a wide variety of problems across robotics, such as Structure from Motion (SfM), Simultaneous Localization and Mapping (SLAM) and calibration. Typically, at their core, they have an optimization problem whose terms only depend on a small subset of variables. Factor graph solvers exploit the locality of problems to drastically reduce the computational time of the Iterative Least-Squares (ILS) methodology. Although extremely powerful, their application is usually limited to unconstrained problems. In this paper, we model constraints over variables within factor graphs by introducing a factor graph version of the method of Lagrange Multipliers. We show the potential of our method by presenting a full navigation stack based on factor graphs. Differently from standard navigation stacks, we can model both optimal control for local planning and localization with factor graphs, and solve the two problems using the standard ILS methodology. We validate our approach in real-world autonomous navigation scenarios, comparing it with the de facto standard navigation stack implemented in ROS. Comparative experiments show that for the application at hand our system outperforms the standard nonlinear programming solver Interior-Point Optimizer (IPOPT) in runtime, while achieving similar solutions.

HiPE: Hierarchical Initialization for Pose Graphs

Pose graph optimization is a non-convex optimization problem encountered in many areas of robotics perception. Its convergence to an accurate solution is conditioned by two factors: the non-linearity of the cost function in use and the initial configuration of the pose variables. In this paper, we present HiPE, a novel hierarchical algorithm for pose graph initialization. Our approach exploits a coarse-grained graph that encodes an abstract representation of the problem geometry. We construct this graph by combining maximum likelihood estimates coming from local regions of the input. By leveraging the sparsity of this representation, we can initialize the pose graph in a non-linear fashion, without computational overhead compared to existing methods. The resulting initial guess can effectively bootstrap the fine-grained optimization that is used to obtain the final solution. In addition, we perform an empirical analysis on the impact of different cost functions on the final estimate. Our experimental evaluation shows that the usage of HiPE leads to a more efficient and robust optimization process, comparing favorably with state-of-the-art methods.