Advancing Myopia To Holism: Fully Contrastive Language-Image Pre-training

In rapidly evolving field of vision-language models (VLMs), contrastive language-image pre-training (CLIP) has made significant strides, becoming foundation for various downstream tasks. However, relying on one-to-one (image, text) contrastive paradigm to learn alignment from large-scale messy web data, CLIP faces a serious myopic dilemma, resulting in biases towards monotonous short texts and shallow visual expressivity. To overcome these issues, this paper advances CLIP into one novel holistic paradigm, by updating both diverse data and alignment optimization. To obtain colorful data with low cost, we use image-to-text captioning to generate multi-texts for each image, from multiple perspectives, granularities, and hierarchies. Two gadgets are proposed to encourage textual diversity. To match such (image, multi-texts) pairs, we modify the CLIP image encoder into multi-branch, and propose multi-to-multi contrastive optimization for image-text part-to-part matching. As a result, diverse visual embeddings are learned for each image, bringing good interpretability and generalization. Extensive experiments and ablations across over ten benchmarks indicate that our holistic CLIP significantly outperforms existing myopic CLIP, including image-text retrieval, open-vocabulary classification, and dense visual tasks.

Points2Plans: From Point Clouds to Long-Horizon Plans with Composable Relational Dynamics

We present Points2Plans, a framework for composable planning with a relational dynamics model that enables robots to solve long-horizon manipulation tasks from partial-view point clouds. Given a language instruction and a point cloud of the scene, our framework initiates a hierarchical planning procedure, whereby a language model generates a high-level plan and a sampling-based planner produces constraint-satisfying continuous parameters for manipulation primitives sequenced according to the high-level plan. Key to our approach is the use of a relational dynamics model as a unifying interface between the continuous and symbolic representations of states and actions, thus facilitating language-driven planning from high-dimensional perceptual input such as point clouds. Whereas previous relational dynamics models require training on datasets of multi-step manipulation scenarios that align with the intended test scenarios, Points2Plans uses only single-step simulated training data while generalizing zero-shot to a variable number of steps during real-world evaluations. We evaluate our approach on tasks involving geometric reasoning, multi-object interactions, and occluded object reasoning in both simulated and real-world settings. Results demonstrate that Points2Plans offers strong generalization to unseen long-horizon tasks in the real world, where it solves over 85% of evaluated tasks while the next best baseline solves only 50%. Qualitative demonstrations of our approach operating on a mobile manipulator platform are made available at sites.google.com/stanford.edu/points2plans.

Low-Complexity Joint Azimuth-Range-Velocity Estimation for Integrated Sensing and Communication with OFDM Waveform

Integrated sensing and communication (ISAC) is a main application scenario of the sixth-generation mobile communication systems. Due to the fast-growing number of antennas and subcarriers in cellular systems, the computational complexity of joint azimuth-range-velocity estimation (JARVE) in ISAC systems is extremely high. This paper studies the JARVE problem for a monostatic ISAC system with orthogonal frequency division multiplexing (OFDM) waveform, in which a base station receives the echos of its transmitted cellular OFDM signals to sense multiple targets. The Cramer-Rao bounds are first derived for JARVE. A low-complexity algorithm is further designed for super-resolution JARVE, which utilizes the proposed iterative subspace update scheme and Levenberg-Marquardt optimization method to replace the exhaustive search of spatial spectrum in multiple-signal-classification (MUSIC) algorithm. Finally, with the practical parameters of 5G New Radio, simulation results verify that the proposed algorithm can reduce the computational complexity by three orders of magnitude and two orders of magnitude compared to the existing three-dimensional MUSIC algorithm and estimation-of-signal-parameters-using-rotational-invariance-techniques (ESPRIT) algorithm, respectively, and also improve the estimation performance.

Fourier Transform-based Wavenumber Domain 3D Imaging in RIS-aided Communication Systems

Radio imaging is rapidly gaining prominence in the design of future communication systems, with the potential to utilize reconfigurable intelligent surfaces (RISs) as imaging apertures. Although the sparsity of targets in three-dimensional (3D) space has led most research to adopt compressed sensing (CS)-based imaging algorithms, these often require substantial computational and memory burdens. Drawing inspiration from conventional Fourier transform (FT)-based imaging methods, our research seeks to accelerate radio imaging in RIS-aided communication systems. To begin, we introduce a two-stage wavenumber domain 3D imaging technique: first, we modify RIS phase shifts to recover the equivalent channel response from the user equipment to the RIS array, subsequently employing traditional FT-based wavenumber domain methods to produce target images. We also determine the diffraction resolution limits of the system through k-space analysis, taking into account factors including system bandwidth, transmission direction, operating frequency, and the angle subtended by the RIS. Addressing the challenge of limited pilots in communication systems, we unveil an innovative algorithm that merges the strengths of both FT- and CS-based techniques by substituting the expansive sensing matrix with FT-based operators. Our simulation outcomes confirm that our proposed FT-based methods achieve high-quality images while demanding few time, memory, and communication resources.

RIS-aided Single-frequency 3D Imaging by Exploiting Multi-view Image Correlations

Retrieving range information in three-dimensional (3D) radio imaging is particularly challenging due to the limited communication bandwidth and pilot resources. To address this issue, we consider a reconfigurable intelligent surface (RIS)-aided uplink communication scenario, generating multiple measurements through RIS phase adjustment. This study successfully realizes 3D single-frequency imaging by exploiting the near-field multi-view image correlations deduced from user mobility. We first highlight the significance of considering anisotropy in multi-view image formation by investigating radar cross-section properties and diffraction resolution limits. We then propose a novel model for joint multi-view 3D imaging that incorporates occlusion effects and anisotropic scattering. These factors lead to slow image support variation and smooth coefficient evolution, which are mathematically modeled as Markov processes. Based on this model, we employ the Expectation Maximization-Turbo-Generalized Approximate Message Passing algorithm for joint multi-view single-frequency 3D imaging with limited measurements. Simulation results reveal the superiority of joint multi-view imaging in terms of enhanced imaging ranges, accuracies, and anisotropy characterization compared to single-view imaging. Combining adjacent observations for joint multi-view imaging enables a reduction in the measurement overhead by 80%.

Joint Range-Velocity-Azimuth Estimation for OFDM-Based Integrated Sensing and Communication

Orthogonal frequency division multiplexing (OFDM)-based integrated sensing and communication (ISAC) is promising for future sixth-generation mobile communication systems. Existing works focus on the joint estimation of the targets' range and velocity for OFDM-based ISAC systems. In contrast, this paper studies the three-dimensional joint estimation (3DJE) of range, velocity, and azimuth for OFDM-based ISAC systems with multiple receive antennas. First, we establish the signal model and derive the Cramer-Rao bounds (CRBs) on the 3DJE. Furthermore, an auto-paired super-resolution 3DJE algorithm is proposed by exploiting the reconstructed observation sub-signal's translational invariance property in the time, frequency, and space domains. Finally, with the 5G New Radio parameter setup, simulation results show that the proposed algorithm achieves better estimation performance and its root mean square error is closer to the root of CRBs than existing methods.

Out of Sight, Still in Mind: Reasoning and Planning about Unobserved Objects with Video Tracking Enabled Memory Models

Robots need to have a memory of previously observed, but currently occluded objects to work reliably in realistic environments. We investigate the problem of encoding object-oriented memory into a multi-object manipulation reasoning and planning framework. We propose DOOM and LOOM, which leverage transformer relational dynamics to encode the history of trajectories given partial-view point clouds and an object discovery and tracking engine. Our approaches can perform multiple challenging tasks including reasoning with occluded objects, novel objects appearance, and object reappearance. Throughout our extensive simulation and real-world experiments, we find that our approaches perform well in terms of different numbers of objects and different numbers of distractor actions. Furthermore, we show our approaches outperform an implicit memory baseline.

Joint Localization and Environment Sensing by Harnessing NLOS Components in RIS-aided mmWave Communication Systems

This study explores the use of non-line-of-sight (NLOS) components in millimeter-wave (mmWave) communication systems for joint localization and environment sensing. The radar cross section (RCS) of a reconfigurable intelligent surface (RIS) is calculated to develop a general path gain model for RISs and traditional scatterers. The results show that RISs have a greater potential to assist in localization due to their ability to maintain high RCSs and create strong NLOS links. A one-stage linear weighted least squares estimator is proposed to simultaneously determine user equipment (UE) locations, velocities, and scatterer (or RIS) locations using line-of-sight (LOS) and NLOS paths. The estimator supports environment sensing and UE localization even using only NLOS paths. A second-stage estimator is also introduced to improve environment sensing accuracy by considering the nonlinear relationship between UE and scatterer locations. Simulation results demonstrate the effectiveness of the proposed estimators in rich scattering environments and the benefits of using NLOS paths for improving UE location accuracy and assisting in environment sensing. The effects of RIS number, size, and deployment on localization performance are also analyzed.

Latent Space Planning for Multi-Object Manipulation with Environment-Aware Relational Classifiers

Objects rarely sit in isolation in everyday human environments. If we want robots to operate and perform tasks in our human environments, they must understand how the objects they manipulate will interact with structural elements of the environment for all but the simplest of tasks. As such, we'd like our robots to reason about how multiple objects and environmental elements relate to one another and how those relations may change as the robot interacts with the world. We examine the problem of predicting inter-object and object-environment relations between previously unseen objects and novel environments purely from partial-view point clouds. Our approach enables robots to plan and execute sequences to complete multi-object manipulation tasks defined from logical relations. This removes the burden of providing explicit, continuous object states as goals to the robot. We explore several different neural network architectures for this task. We find the best performing model to be a novel transformer-based neural network that both predicts object-environment relations and learns a latent-space dynamics function. We achieve reliable sim-to-real transfer without any fine-tuning. Our experiments show that our model understands how changes in observed environmental geometry relate to semantic relations between objects. We show more videos on our website: https://sites.google.com/view/erelationaldynamics.

Planning for Multi-Object Manipulation with Graph Neural Network Relational Classifiers

Objects rarely sit in isolation in human environments. As such, we'd like our robots to reason about how multiple objects relate to one another and how those relations may change as the robot interacts with the world. To this end, we propose a novel graph neural network framework for multi-object manipulation to predict how inter-object relations change given robot actions. Our model operates on partial-view point clouds and can reason about multiple objects dynamically interacting during the manipulation. By learning a dynamics model in a learned latent graph embedding space, our model enables multi-step planning to reach target goal relations. We show our model trained purely in simulation transfers well to the real world. Our planner enables the robot to rearrange a variable number of objects with a range of shapes and sizes using both push and pick and place skills.