Learning from demonstration is a powerful method for robotic skill acquisition. Nevertheless, a critical limitation lies in the substantial costs associated with gathering demonstration datasets, typically action-labeled robot data, which creates a fundamental constraint in the field. Video data offer a compelling solution as an alternative rich data source, containing diverse behavioral and physical knowledge. This study introduces G3M, an innovative framework that exploits video data viaGraph-to-GraphsGenerativeModeling, which pre-trains models to generate future graphs conditioned on the graph within a video frame. The proposed G3M abstracts video frame into graph representations by identifying object and visual action vertices for capturing state information. It then effectively models internal structures and spatial relationships present in these graph constructions, with the objective of predicting forthcoming graphs. The generated graphs function as conditional inputs that guide the control policy in determining robotic behaviors. This concise method effectively encodes critical spatial relationships while facilitating accurate prediction of subsequent graph sequences, thus allowing the development of resilient control policy despite constraints in action-annotated training samples. Furthermore, these transferable graph representations enable the effective extraction of manipulation knowledge through human videos as well as recordings from robots with different embodiments. The experimental results demonstrate that G3M attains superior performance using merely 20% action-labeled data relative to comparable approaches. Moreover, our method outperforms the state-of-the-art method, showing performance gains exceeding 19% in simulated environments and 23% in real-world experiments, while delivering improvements of over 35% in cross-embodiment transfer experiments and exhibiting strong performance on long-horizon tasks. Our project page is available athttps://g3m-project.github.io/.

Learning from demonstration is a powerful method for robotic skill acquisition. However, the significant expense of collecting such action-labeled robot data presents a major bottleneck. Video data, a rich data source encompassing diverse behavioral and physical knowledge, emerges as a promising alternative. In this paper, we present GraphMimic, a novel paradigm that leverages video data via graph-to-graphs generative modeling, which pre-trains models to generate future graphs conditioned on the graph within a video frame. Specifically, GraphMimic abstracts video frames into object and visual action vertices, and constructs graphs for state representations. The graph generative modeling network then effectively models internal structures and spatial relationships within the constructed graphs, aiming to generate future graphs. The generated graphs serve as conditions for the control policy, mapping to robot actions. Our concise approach captures important spatial relations and enhances future graph generation accuracy, enabling the acquisition of robust policies from limited action-labeled data. Furthermore, the transferable graph representations facilitate the effective learning of manipulation skills from cross-embodiment videos. Our experiments exhibit that GraphMimic achieves superior performance using merely 20% action-labeled data. Moreover, our method outperforms the state-of-the-art method by over 17% and 23% in simulation and real-world experiments, and delivers improvements of over 33% in cross-embodiment transfer experiments.

In various robotic applications, understanding accurate object poses for robots is essential for high-precision tasks such as factory assembly or daily insertions. Tactile sensing, which compensates for visual information, offers rich texture-based or force-based data for object pose estimation. However, previous methods for pose estimation typically over-look dynamic situations, such as slippage of grasped objects or movement of contacted objects during interactions with the environment, thus increasing the complexity of pose estimation. To address these challenges, we propose an efficient method that utilizes visual and tactile sensing to estimate object poses through particle filtering. We leverage visual information to track the pose of the contacted object in real-time and estimate the pose changes of the grasped object using displacement data obtained from tactile sensors. Our experimental evaluation on 13 objects with diverse geometric shapes demonstrated the ability to estimate high-precision poses, which revealed the robot's powerful ability to cope with dynamic scenes for compelled motion of objects, proving our framework's adaptability in practical scenarios with uncertainty.

Image feature matching is a fundamental task in computer vision. Existing local feature matching methods can establish robust correspondences between image pairs. However, these methods heavily rely on dense local image features, making them susceptible to significant perspective differences, characterized by rotation and scale changes. To alleviate this limitation, we introduce a novel oriented Overlapping Region Alignment method, named ORA-NET, which presents a concise and efficient approach to enhance the performance of image feature matching methods. We introduce the Multidirectional Cross-scale Feature Aggregation module to aggregate rotation-equivariant features across multiple scales and model long-range dependencies. Additionally, the Oriented Overlap Alignment module estimates scale and rotation differences within overlapping regions using a coarse-to-fine rotation correction approach. Importantly, our method serves as a plug-and-play module that can be seamlessly integrated into other correspondence matching pipelines. Experimental results demonstrate that ORA-NET significantly enhances the matching performance of existing local feature matching methods, particularly in scenarios involving substantial perspective differences.

Learning from human demonstrations is an emerging trend for designing intelligent robotic systems. However, previous methods typically regard videos as instructions, simply dividing videos into action sequences for robotic repetition, which pose obstacles to generalization to diverse tasks or object instances. In this paper, we propose a different perspective, considering human demonstration videos not as mere instructions, but as a source of knowledge for robots. Motivated by this perspective and the remarkable comprehension and generalization capabilities exhibited by large language models (LLMs), we propose DigKnow, a method that DIstills Generalizable KNOWledge with a hierarchical structure. Specifically, DigKnow begins by converting human demonstration video frames into observation knowledge. This knowledge is then subjected to analysis to extract human action knowledge and further distilled into pattern knowledge that comprises task and object instances, resulting in the acquisition of generalizable knowledge with a hierarchical structure. In settings with different tasks or object instances, DigKnow retrieves relevant knowledge for the current task and object instances. Subsequently, the LLM-based planner conducts planning based on the retrieved knowledge, and the policy executes actions in line with the plan to achieve the designated task. Utilizing the retrieved knowledge, we validate and rectify planning and execution outcomes, resulting in a substantial enhancement of the success rate. Experimental results across a range of tasks and scenes demonstrate the effectiveness of this approach in facilitating real-world robots to accomplish tasks with the knowledge derived from human demonstrations.

The ability to perform generalizable and precise grasping on functional object parts is a prerequisite for robotic manipulation in open environments. Recent foundation models have demonstrated promising semantic correspondence capabilities in guiding robots to grasp similar parts across objects with resembling shapes and poses. However, existing works struggle to generalize precise grasp poses when the target objects exhibit substantial geometric and positional variations. To tackle this challenge, we present PartGrasp, a method that achieves precise part grasping through hierarchical integration of highly generalizable semantic correspondence and precise geometric registration. Specifically, we first build a grasp knowledge bank by extracting grasp poses and object meshes from demonstrations. Upon retrieving a reference from this bank, we initially perform a coarse alignment using semantic correspondence, followed by a fine registration that adapts to geometric variations. This approach achieves fine-grained generalization of part grasping that is robust to both shape and pose variations. Extensive experiments demonstrate the efficacy of our method in terms of both generalization capability and accuracy. Videos and more details are available on our project site: https://part-grasp.github.io/partgrasp/.

The ability to perform reliable long-horizon task planning is crucial for deploying robots in real-world environments. However, directly employing Large Language Models (LLMs) as action sequence generators often results in low success rates due to their limited reasoning ability for long-horizon embodied tasks. In the STEP framework, we construct a subgoal tree through a pair of closed-loop models: a subgoal decomposition model and a leaf node termination model. Within this framework, we develop a hierarchical tree structure that spans from coarse to fine resolutions. The subgoal decomposition model leverages a foundation LLM to break down complex goals into manageable subgoals, thereby spanning the subgoal tree. The leaf node termination model provides real-time feedback based on environmental states, determining when to terminate the tree spanning and ensuring each leaf node can be directly converted into a primitive action. Experiments conducted in both the VirtualHome WAH-NL benchmark and on real robots demonstrate that STEP achieves long-horizon embodied task completion with success rates up to 34% (WAH-NL) and 25% (real robot) outperforming SOTA methods.

Video data provides an accessible and rich source beyond expensive action-labeled robot data for advancing robotic learning paradigms. Motivated by this potential, researchers investigate methods to exploit video data in robotic learning. Recent approaches can be primarily divided into two categories: Action-based approaches tokenize latent actions from videos for policy pre-training. State-based approaches pre-train models to predict subsequent states. The former establishes rich motion priors, while the latter empowers the robot to anticipate future events. These complementary capabilities suggest significant potential for integration into a unified framework. In this paper, we propose UniMimic, a novel approach unifying latent action and latent state pre-training from videos. We first train a unified tokenizer to learn latent states from video frames while deriving latent actions between state tokens. Subsequently, the policy is pre-trained on videos to predict these latent actions and subsequent latent states. Finally, the policy is fine-tuned on an action-labeled robot dataset to transfer the learned priors to precise robot execution. Experiments exhibit that our pre-training stage enhances the performance by 19% in the Libero benchmark and improves the average number of tasks completed in a row of 5 from 2.50 and 2.35 to 3.89 and 3.73 in the CALVIN benchmark. In the real-world experiments, our method still delivers improvements exceeding 36%.

Point cloud registration is a fundamental task in the fields of computer vision and robotics. Recent advancements in transformer-based methods have demonstrated enhanced performance in this domain. However, the standard attention mechanisms employed in these approaches tend to incorporate numerous points of low relevance, and therefore struggle to focus their attention weights on sparse yet meaningful points. This inefficiency leads to limited local structure modeling capabilities and quadratic computational complexity. To overcome these limitations, we propose the Point Tree Transformer (PTT), a novel transformer-based approach for point cloud registration that efficiently extracts comprehensive local and global features while maintaining linear computational complexity. The PTT constructs hierarchical feature trees from point clouds in a coarse-to-dense manner, and introduces a novel Point Tree Attention (PTA) mechanism. This mechanism adheres to the tree structure to facilitate the progressive convergence of attended regions toward salient points. Specifically, each tree layer selectively identifies a subset of relevant points with the highest attention scores, and subsequent layers focus attention on areas of significant relevance, derived from the child points of the selected point set. The feature extraction process additionally incorporates coarse point features that capture high-level semantic information, thus facilitating local structure modeling and the progressive integration of multiscale information. Consequently, the PTA enables the model to focus on essential local structures and extract intricate local information while maintaining linear computational complexity. Extensive experiments conducted on the 3DMatch, ModelNet40, and KITTI datasets demonstrate that our method outperforms state-of-the-art methods in terms of performance. The code for our method is publicly available at https://github.com/CGuangyan-BIT/PTT.

Point cloud registration is essential in computer vision and robotics. Recently, transformer-based methods have achieved advanced point cloud registration performance. However, the standard attention mechanism utilized in these methods considers many low-relevance points, and it has difficulty focusing its attention weights on sparse and meaningful points, leading to limited local structure modeling capabilities and quadratic computational complexity. To address these limitations, we present the Tree-based Transformer (TrT), which is able to extract abundant local and global features with linear computational complexity. Specifically, the TrT builds coarse-to-dense feature trees, and a novel Tree-based Attention (TrA) is proposed to guide the progressive convergence of the attended regions toward meaningful points and to structurize point clouds following tree structures. In each layer, the top ${\mathcal{S}}$ key points with the highest attention scores are selected, such that in the next layer, attention is evaluated only within the specified high-relevance regions, corresponding to the child points of these selected ${\mathcal{S}}$ points. Additionally, coarse features containing high-level semantic information are incorporated into the child points to guide the feature extraction process, facilitating local structure modeling and multiscale information integration. Consequently, TrA enables the model to focus on critical local structures and extract rich local information with linear computational complexity. Experiments demonstrate that our method achieves state-of-the-art performance on 3DMatch and KITTI benchmarks. The code for our method is publicly available at https://github.com/CGuangyan-BIT/TrT.

Visual imitation learning (VIL) provides an efficient and intuitive strategy for robotic systems to acquire novel skills. Recent advancements in Vision Language Models (VLMs) have demonstrated remarkable performance in vision and language reasoning capabilities for VIL tasks. Despite the progress, current VIL methods naively employ VLMs to learn high-level plans from human videos, relying on pre-defined motion primitives for executing physical interactions, which remains a major bottleneck. In this work, we present VLMimic, a novel paradigm that harnesses VLMs to directly learn even fine-grained action levels, only given a limited number of human videos. Specifically, VLMimic first grounds object-centric movements from human videos, and learns skills using hierarchical constraint representations, facilitating the derivation of skills with fine-grained action levels from limited human videos. These skills are refined and updated through an iterative comparison strategy, enabling efficient adaptation to unseen environments. Our extensive experiments exhibit that our VLMimic, using only 5 human videos, yields significant improvements of over 27% and 21% in RLBench and real-world manipulation tasks, and surpasses baselines by more than 37% in long-horizon tasks. Code and videos are available on our anonymous homepage.

Point cloud registration is an essential technology in computer vision and robotics. Recently, transformer-based methods have achieved advanced performance in point cloud registration by utilizing the advantages of the transformer in order-invariance and modeling dependencies to aggregate information. However, they still suffer from indistinct feature extraction, sensitivity to noise, and outliers, owing to three major limitations: 1) the adoption of CNNs fails to model global relations due to their local receptive fields, resulting in extracted features susceptible to noise; 2) the shallow-wide architecture of transformers and the lack of positional information lead to indistinct feature extraction due to inefficient information interaction; and 3) the insufficient consideration of geometrical compatibility leads to the ambiguous identification of incorrect correspondences. To address the above-mentioned limitations, a novel full transformer network for point cloud registration is proposed, named the deep interaction transformer (DIT), which incorporates: 1) a point cloud structure extractor (PSE) to retrieve structural information and model global relations with the local feature integrator (LFI) and transformer encoders; 2) a deep-narrow point feature transformer (PFT) to facilitate deep information interaction across a pair of point clouds with positional information, such that transformers establish comprehensive associations and directly learn the relative position between points; and 3) a geometric matching-based correspondence confidence evaluation (GMCCE) method to measure spatial consistency and estimate correspondence confidence by the designed triangulated descriptor. Extensive experiments on the ModelNet40, ScanObjectNN, and 3DMatch datasets demonstrate that our method is capable of precisely aligning point clouds, consequently, achieving superior performance compared with state-of-the-art methods. The code is publicly available at https://github.com/CGuangyan-BIT/DIT.

Point cloud registration is essential in computer vision and robotics. In this paper, a critical observation is made that the invisible parts of each point cloud can be directly utilized as inherent masks, and the aligned point cloud pair can be regarded as the reconstruction target. Motivated by this observation, we rethink the point cloud registration problem as a masking and reconstruction task. To this end, a generic and concise auxiliary training network, the Masked Reconstruction Auxiliary Network (MRA), is proposed. The MRA reconstructs the complete point cloud by separately using the encoded features of each point cloud obtained from the backbone, guiding the contextual features in the backbone to capture fine-grained geometric details and the overall structures of point cloud pairs. Unlike recently developed high-performing methods that incorporate specific encoding methods into transformer models, which sacrifice versatility and introduce significant computational complexity during the inference process, our MRA can be easily inserted into other methods to further improve registration accuracy. Additionally, the MRA is detached after training, thereby avoiding extra computational complexity during the inference process. Building upon the MRA, we present a novel transformer-based method, the Masked Reconstruction Transformer (MRT), which achieves both precise and efficient alignment using standard transformers. Extensive experiments conducted on the 3DMatch, ModelNet40, and KITTI datasets demonstrate the superior performance of our MRT over state-of-the-art methods. Codes are available at https://github.com/CGuangyan-BIT/MRA.

Point cloud registration is a crucial technology in the fields of robotics and computer vision. Despite the significant advances in point cloud registration enabled by Transformer-based methods, limitations persist due to indistinct feature extraction, noise sensitivity, and outlier handling. These limitations stem from three factors: (1) the inefficiency of convolutional neural networks (CNNs) to capture global relationships due to their local receptive fields, resulting in extracted features susceptible to noise; (2) the shallow-wide architecture of Transformers, coupled with a lack of positional information, leading to inefficient information interaction and indistinct feature extraction; and (3) the omission of geometrical compatibility leads to ambiguous identification of incorrect correspondences. To overcome these limitations, we propose the Deep Interactive Full Transformer (DIFT) network for point cloud registration, which consists of three key components: (1) a Point Cloud Structure Extractor (PSE) for modeling global relationships and retrieving structural information; (2) a Point Feature Transformer (PFT) for establishing comprehensive associations and directly learning the relative positions between points; and (3) a Geometric Matching-based Correspondence Confidence Evaluation (GMCCE) method for measuring spatial consistency and estimating correspondence confidence. Experimental results on ModelNet40 and 3DMatch datasets demonstrate the superior performance of our proposed method compared to existing state-of-the-art methods. The code for our method is publicly available at https://github.com/CGuangyan-BIT/DIFT.

Large language models (LLMs) based on the generative pre-training transformer (GPT) have demonstrated remarkable effectiveness across a diverse range of downstream tasks. Inspired by the advancements of the GPT, we present PointGPT, a novel approach that extends the concept of GPT to point clouds, addressing the challenges associated with disorder properties, low information density, and task gaps. Specifically, a point cloud auto-regressive generation task is proposed to pre-train transformer models. Our method partitions the input point cloud into multiple point patches and arranges them in an ordered sequence based on their spatial proximity. Then, an extractor-generator based transformer decode, with a dual masking strategy, learns latent representations conditioned on the preceding point patches, aiming to predict the next one in an auto-regressive manner. To explore scalability and enhance performance, a larger pre-training dataset is collected. Additionally, a subsequent post-pre-training stage is introduced, incorporating a labeled hybrid dataset. Our scalable approach allows for learning high-capacity models that generalize well, achieving state-of-the-art performance on various downstream tasks. In particular, our approach achieves classification accuracies of 94.9% on the ModelNet40 dataset and 93.4% on the ScanObjectNN dataset, outperforming all other transformer models. Furthermore, our method also attains new state-of-the-art accuracies on all four few-shot learning benchmarks. Codes are available at https://github.com/CGuangyan-BIT/PointGPT.