This paper presents the Embedding Pose Graph (EPG), an innovative method that combines the strengths of foundation models with a simple 3D representation suitable for robotics applications. Addressing the need for efficient spatial understanding in robotics, EPG provides a compact yet powerful approach by attaching foundation model features to the nodes of a pose graph. Unlike traditional methods that rely on bulky data formats like voxel grids or point clouds, EPG is lightweight and scalable. It facilitates a range of robotic tasks, including open-vocabulary querying, disambiguation, image-based querying, language-directed navigation, and re-localization in 3D environments. We showcase the effectiveness of EPG in handling these tasks, demonstrating its capacity to improve how robots interact with and navigate through complex spaces. Through both qualitative and quantitative assessments, we illustrate EPG's strong performance and its ability to outperform existing methods in re-localization. Our work introduces a crucial step forward in enabling robots to efficiently understand and operate within large-scale 3D spaces.
Human following is a crucial feature of human-robot interaction, yet it poses numerous challenges to mobile agents in real-world scenarios. Some major hurdles are that the target person may be in a crowd, obstructed by others, or facing away from the agent. To tackle these challenges, we present a novel person re-identification module composed of three parts: a 360-degree visual registration, a neural-based person re-identification using human faces and torsos, and a motion tracker that records and predicts the target person's future position. Our human-following system also addresses other challenges, including identifying fast-moving targets with low latency, searching for targets that move out of the camera's sight, collision avoidance, and adaptively choosing different following mechanisms based on the distance between the target person and the mobile agent. Extensive experiments show that our proposed person re-identification module significantly enhances the human-following feature compared to other baseline variants.
We present a method for generating, predicting, and using Spatiotemporal Occupancy Grid Maps (SOGM), which embed future semantic information of real dynamic scenes. We present an auto-labeling process that creates SOGMs from noisy real navigation data. We use a 3D-2D feedforward architecture, trained to predict the future time steps of SOGMs, given 3D lidar frames as input. Our pipeline is entirely self-supervised, thus enabling lifelong learning for real robots. The network is composed of a 3D back-end that extracts rich features and enables the semantic segmentation of the lidar frames, and a 2D front-end that predicts the future information embedded in the SOGM representation, potentially capturing the complexities and uncertainties of real-world multi-agent, multi-future interactions. We also design a navigation system that uses these predicted SOGMs within planning, after they have been transformed into Spatiotemporal Risk Maps (SRMs). We verify our navigation system's abilities in simulation, validate it on a real robot, study SOGM predictions on real data in various circumstances, and provide a novel indoor 3D lidar dataset, collected during our experiments, which includes our automated annotations.
We present a novel method for generating, predicting, and using Spatiotemporal Occupancy Grid Maps (SOGM), which embed future information of dynamic scenes. Our automated generation process creates groundtruth SOGMs from previous navigation data. We use them to train a 3D-2D feedforward architecture, in a self-supervised fashion, thus enabling lifelong learning for robots. The automated generation process uses ray-tracing to label points in the robot environment based on their dynamic properties. The network is composed of a 3D back-end that extracts rich features and enables the semantic segmentation of the lidar frames, and a 2D front-end that predicts the future information embedded in the SOGMs. We also design a navigation pipeline using these predicted SOGMs. We provide both quantitative and qualitative insights into the predictions and validate our choices of network design with an ablation study.
We present unsupervised parameter learning in a Gaussian variational inference setting that combines classic trajectory estimation for mobile robots with deep learning for rich sensor data, all under a single learning objective. The framework is an extension of an existing system identification method that optimizes for the observed data likelihood, which we improve with modern advances in batch trajectory estimation and deep learning. Though the framework is general to any form of parameter learning and sensor modality, we demonstrate application to feature and uncertainty learning with a deep network for 3D lidar odometry. Our framework learns from only the on-board lidar data, and does not require any form of groundtruth supervision. We demonstrate that our lidar odometry performs better than existing methods that learn the full estimator with a deep network, and comparable to state-of-the-art ICP-based methods on the KITTI odometry dataset. We additionally show results on lidar data from the Oxford RobotCar dataset.
We present a self-supervised learning approach for the semantic segmentation of lidar frames. Our method is used to train a deep point cloud segmentation architecture without any human annotation. The annotation process is automated with the combination of simultaneous localization and mapping (SLAM) and ray-tracing algorithms. By performing multiple navigation sessions in the same environment, we are able to identify permanent structures, such as walls, and disentangle short-term and long-term movable objects, such as people and tables, respectively. New sessions can then be performed using a network trained to predict these semantic labels. We demonstrate the ability of our approach to improve itself over time, from one session to the next. With semantically filtered point clouds, our robot can navigate through more complex scenarios, which, when added to the training pool, help to improve our network predictions. We provide insights into our network predictions and show that our approach can also improve the performances of common localization techniques.
Recent attempts at introducing rotation invariance or equivariance in 3D deep learning approaches have shown promising results, but these methods still struggle to reach the performances of standard 3D neural networks. In this work we study the relation between equivariance and invariance in 3D point convolutions. We show that using rotation-equivariant alignments, it is possible to make any convolutional layer rotation-invariant. Furthermore, we improve this simple alignment procedure by using the alignment themselves as features in the convolution, and by combining multiple alignments together. With this core layer, we design rotation-invariant architectures which improve state-of-the-art results in both object classification and semantic segmentation and reduces the gap between rotation-invariant and standard 3D deep learning approaches.
We present Kernel Point Convolution (KPConv), a new design of point convolution, i.e. that operates on point clouds without any intermediate representation. The convolution weights of KPConv are located in Euclidean space by kernel points, and applied to the input points close to them. Its capacity to use any number of kernel points gives KPConv more flexibility than fixed grid convolutions. Furthermore, these locations are continuous in space and can be learned by the network. Therefore, KPConv can be extended to deformable convolutions that learn to adapt kernel points to local geometry. Thanks to a regular subsampling strategy, KPConv is also efficient and robust to varying densities. Whether they use deformable KPConv for complex tasks, or rigid KPconv for simpler tasks, our networks outperform state-of-the-art classification and segmentation approaches on several datasets. We also offer ablation studies and visualizations to provide understanding of what has been learned by KPConv and to validate the descriptive power of deformable KPConv.
This paper introduces a new definition of multiscale neighborhoods in 3D point clouds. This definition, based on spherical neighborhoods and proportional subsampling, allows the computation of features with a consistent geometrical meaning, which is not the case when using k-nearest neighbors. With an appropriate learning strategy, the proposed features can be used in a random forest to classify 3D points. In this semantic classification task, we show that our multiscale features outperform state-of-the-art features using the same experimental conditions. Furthermore, their classification power competes with more elaborate classification approaches including Deep Learning methods.