Abstract:The vanilla autoregressive image generation model generates visual tokens in a step-by-step fashion, which limits the ability to capture holistic relationships among token sequences. Moreover, most visual tokenizers map local image patches into latent tokens, leading to limited global information. To address this, we introduce \textit{Hita}, a novel image tokenizer for autoregressive (AR) image generation. It introduces a holistic-to-local tokenization scheme with learnable holistic queries and local patch tokens. Besides, Hita incorporates two key strategies for improved alignment with the AR generation process: 1) it arranges a sequential structure with holistic tokens at the beginning followed by patch-level tokens while using causal attention to maintain awareness of previous tokens; and 2) before feeding the de-quantized tokens into the decoder, Hita adopts a lightweight fusion module to control information flow to prioritize holistic tokens. Extensive experiments show that Hita accelerates the training speed of AR generators and outperforms those trained with vanilla tokenizers, achieving \textbf{2.59 FID} and \textbf{281.9 IS} on the ImageNet benchmark. A detailed analysis of the holistic representation highlights its ability to capture global image properties such as textures, materials, and shapes. Additionally, Hita also demonstrates effectiveness in zero-shot style transfer and image in-painting. The code is available at \href{https://github.com/CVMI-Lab/Hita}{https://github.com/CVMI-Lab/Hita}
Abstract:Vision-Language-Action (VLA) models have recently made significant advance in multi-task, end-to-end robotic control, due to the strong generalization capabilities of Vision-Language Models (VLMs). A fundamental challenge in developing such models is effectively aligning the vision-language space with the robotic action space. Existing approaches typically rely on directly fine-tuning VLMs using expert demonstrations. However, this strategy suffers from a spatio-temporal gap, resulting in considerable data inefficiency and heavy reliance on human labor. Spatially, VLMs operate within a high-level semantic space, whereas robotic actions are grounded in low-level 3D physical space; temporally, VLMs primarily interpret the present, while VLA models anticipate future actions. To overcome these challenges, we propose a novel training paradigm, ROSA, which leverages robot state estimation to improve alignment between vision-language and action spaces. By integrating robot state estimation data obtained via an automated process, ROSA enables the VLA model to gain enhanced spatial understanding and self-awareness, thereby boosting performance and generalization. Extensive experiments in both simulated and real-world environments demonstrate the effectiveness of ROSA, particularly in low-data regimes.
Abstract:Embodied 3D grounding aims to localize target objects described in human instructions from ego-centric viewpoint. Most methods typically follow a two-stage paradigm where a trained 3D detector's optimized backbone parameters are used to initialize a grounding model. In this study, we explore a fundamental question: Does embodied 3D grounding benefit enough from detection? To answer this question, we assess the grounding performance of detection models using predicted boxes filtered by the target category. Surprisingly, these detection models without any instruction-specific training outperform the grounding models explicitly trained with language instructions. This indicates that even category-level embodied 3D grounding may not be well resolved, let alone more fine-grained context-aware grounding. Motivated by this finding, we propose DEGround, which shares DETR queries as object representation for both DEtection and Grounding and enables the grounding to benefit from basic category classification and box detection. Based on this framework, we further introduce a regional activation grounding module that highlights instruction-related regions and a query-wise modulation module that incorporates sentence-level semantic into the query representation, strengthening the context-aware understanding of language instructions. Remarkably, DEGround outperforms state-of-the-art model BIP3D by 7.52\% at overall accuracy on the EmbodiedScan validation set. The source code will be publicly available at https://github.com/zyn213/DEGround.
Abstract:In this paper, we propose PADriver, a novel closed-loop framework for personalized autonomous driving (PAD). Built upon Multi-modal Large Language Model (MLLM), PADriver takes streaming frames and personalized textual prompts as inputs. It autoaggressively performs scene understanding, danger level estimation and action decision. The predicted danger level reflects the risk of the potential action and provides an explicit reference for the final action, which corresponds to the preset personalized prompt. Moreover, we construct a closed-loop benchmark named PAD-Highway based on Highway-Env simulator to comprehensively evaluate the decision performance under traffic rules. The dataset contains 250 hours videos with high-quality annotation to facilitate the development of PAD behavior analysis. Experimental results on the constructed benchmark show that PADriver outperforms state-of-the-art approaches on different evaluation metrics, and enables various driving modes.
Abstract:The rapid development of Large Multimodal Models (LMMs) for 2D images and videos has spurred efforts to adapt these models for interpreting 3D scenes. However, the absence of large-scale 3D vision-language datasets has posed a significant obstacle. To address this issue, typical approaches focus on injecting 3D awareness into 2D LMMs by designing 3D input-level scene representations. This work provides a new perspective. We introduce reconstructive visual instruction tuning with 3D-awareness (Ross3D), which integrates 3D-aware visual supervision into the training procedure. Specifically, it incorporates cross-view and global-view reconstruction. The former requires reconstructing masked views by aggregating overlapping information from other views. The latter aims to aggregate information from all available views to recover Bird's-Eye-View images, contributing to a comprehensive overview of the entire scene. Empirically, Ross3D achieves state-of-the-art performance across various 3D scene understanding benchmarks. More importantly, our semi-supervised experiments demonstrate significant potential in leveraging large amounts of unlabeled 3D vision-only data.
Abstract:Multi-hand semantic grasp generation aims to generate feasible and semantically appropriate grasp poses for different robotic hands based on natural language instructions. Although the task is highly valuable, due to the lack of multi-hand grasp datasets with fine-grained contact description between robotic hands and objects, it is still a long-standing difficult task. In this paper, we present Multi-GraspSet, the first large-scale multi-hand grasp dataset with automatically contact annotations. Based on Multi-GraspSet, we propose Multi-GraspLLM, a unified language-guided grasp generation framework. It leverages large language models (LLM) to handle variable-length sequences, generating grasp poses for diverse robotic hands in a single unified architecture. Multi-GraspLLM first aligns the encoded point cloud features and text features into a unified semantic space. It then generates grasp bin tokens which are subsequently converted into grasp pose for each robotic hand via hand-aware linear mapping. The experimental results demonstrate that our approach significantly outperforms existing methods on Multi-GraspSet. More information can be found on our project page https://multi-graspllm.github.io.
Abstract:Generating high-fidelity, controllable, and annotated training data is critical for autonomous driving. Existing methods typically generate a single data form directly from a coarse scene layout, which not only fails to output rich data forms required for diverse downstream tasks but also struggles to model the direct layout-to-data distribution. In this paper, we introduce UniScene, the first unified framework for generating three key data forms - semantic occupancy, video, and LiDAR - in driving scenes. UniScene employs a progressive generation process that decomposes the complex task of scene generation into two hierarchical steps: (a) first generating semantic occupancy from a customized scene layout as a meta scene representation rich in both semantic and geometric information, and then (b) conditioned on occupancy, generating video and LiDAR data, respectively, with two novel transfer strategies of Gaussian-based Joint Rendering and Prior-guided Sparse Modeling. This occupancy-centric approach reduces the generation burden, especially for intricate scenes, while providing detailed intermediate representations for the subsequent generation stages. Extensive experiments demonstrate that UniScene outperforms previous SOTAs in the occupancy, video, and LiDAR generation, which also indeed benefits downstream driving tasks.
Abstract:Existing policy learning methods predominantly adopt the task-centric paradigm, necessitating the collection of task data in an end-to-end manner. Consequently, the learned policy tends to fail to tackle novel tasks. Moreover, it is hard to localize the errors for a complex task with multiple stages due to end-to-end learning. To address these challenges, we propose RoboMatrix, a skill-centric and hierarchical framework for scalable task planning and execution. We first introduce a novel skill-centric paradigm that extracts the common meta-skills from different complex tasks. This allows for the capture of embodied demonstrations through a kill-centric approach, enabling the completion of open-world tasks by combining learned meta-skills. To fully leverage meta-skills, we further develop a hierarchical framework that decouples complex robot tasks into three interconnected layers: (1) a high-level modular scheduling layer; (2) a middle-level skill layer; and (3) a low-level hardware layer. Experimental results illustrate that our skill-centric and hierarchical framework achieves remarkable generalization performance across novel objects, scenes, tasks, and embodiments. This framework offers a novel solution for robot task planning and execution in open-world scenarios. Our software and hardware are available at https://github.com/WayneMao/RoboMatrix.
Abstract:Efficient acquisition of real-world embodied data has been increasingly critical. However, large-scale demonstrations captured by remote operation tend to take extremely high costs and fail to scale up the data size in an efficient manner. Sampling the episodes under a simulated environment is a promising way for large-scale collection while existing simulators fail to high-fidelity modeling on texture and physics. To address these limitations, we introduce the RoboGSim, a real2sim2real robotic simulator, powered by 3D Gaussian Splatting and the physics engine. RoboGSim mainly includes four parts: Gaussian Reconstructor, Digital Twins Builder, Scene Composer, and Interactive Engine. It can synthesize the simulated data with novel views, objects, trajectories, and scenes. RoboGSim also provides an online, reproducible, and safe evaluation for different manipulation policies. The real2sim and sim2real transfer experiments show a high consistency in the texture and physics. Moreover, the effectiveness of synthetic data is validated under the real-world manipulated tasks. We hope RoboGSim serves as a closed-loop simulator for fair comparison on policy learning. More information can be found on our project page https://robogsim.github.io/ .
Abstract:Instruction tuning constitutes a prevalent technique for tailoring Large Vision Language Models (LVLMs) to meet individual task requirements. To date, most of the existing approaches are confined to single-task adaptation, whereas the requirements in real-world scenarios are inherently varied and continually evolving. Thus an ideal LVLM should sustain continual instruction tuning in the face of stream-task distributions (i.e., different domains, emerging capabilities, and new datasets) while minimizing the forgetting of previously acquired knowledge. To achieve this, we propose a new benchmark for COntinuAl inStruction Tuning on LVLMs (COAST), which encompasses the aforementioned domain-incremental, capability-incremental, and dataset-incremental configurations. In terms of methodology, we propose Continual LLaVA, a rehearsal-free method tailored for continual instruction tuning in LVLMs. To circumvent the additional overhead associated with experience replay, we freeze LVLMs and construct the dual increment embeddings for each input instruction to facilitate parameter-efficient tuning. Specifically, the increment embeddings can be decomposed into two principal components: 1) intrinsic increment embeddings to encode task-specific characteristics. To achieve this, we set up a low-rank pool containing candidate embeddings, from which we select the relevant ones based on their similarity with the user instructions; 2) contextual increment embeddings to investigate the inter-dependencies across tasks. In this regard, the low-rank embeddings chosen in the previous tasks are aggregated via learnable weighted sum to provide complementary hints. Extensive experiments indicate that the proposed Continual LLaVA outperforms previous methods by significantly reducing the forgetting during the continual instruction tuning process.