The development of embodied AI (EAI) critically depends on efficient data acquisition, yet faces persistent challenges including high costs, limited training scenarios, and lack of standardized datasets. We present AIRSPEED, an open source data production platform designed to address these bottlenecks through three core innovations. First, AIRSPEED achieves hardware–software decoupling via unified robot and simulation interfaces, enabling seamless integration with diverse data collection devices and simulation platforms. Second, it supports comprehensive data production methods spanning teleoperation and teaching approaches, as well as synthetic data generation through data synthesis and virtual teleoperation. Third, AIRSPEED automates pyramid-structured dataset construction compatible with both HDF5 and LeRobot formats, significantly reducing manual overhead. Experimental validation demonstrates substantial efficiency gains, achieving up to 35.6× acceleration in dataset construction and 6.0× overall speedup compared to manual workflows. With end-to-end latency as low as 3 ms and compression throughput exceeding 296 MB/s, AIRSPEED establishes a scalable foundation for EAI data production. AISPEED is open sourced on this website: URL .

Learning robotic skills from raw human videos remains a non-trivial challenge. Previous works tackled this problem by leveraging behavior cloning or learning reward functions from videos. Despite their remarkable performances, they may introduce several issues, such as the necessity for robot actions, requirements for consistent viewpoints and similar layouts between human and robot videos, as well as low sample efficiency. To this end, our key insight is to learn task priors by contrasting videos and to learn action priors through imitating trajectories from videos, and to utilize the task priors to guide trajectories to adapt to novel scenarios. We propose a three-stage skill learning framework denoted as Contrast-Imitate-Adapt (CIA). An interaction-aware alignment transformer is proposed to learn task priors by temporally aligning video pairs. Then a trajectory generation model is used to learn action priors. To adapt to novel scenarios different from human videos, the Inversion-Interaction method is designed to initialize coarse trajectories and refine them by limited interaction. In addition, CIA introduces an optimization method based on semantic directions of trajectories for interaction security and sample efficiency. The alignment distances computed by IAAformer are used as the rewards. We evaluate CIA in six real-world everyday tasks, and empirically demonstrate that CIA significantly outperforms previous state-of-the-art works in terms of task success rate and generalization to diverse novel scenarios layouts and object instances.Note to Practitioners—This work aims to study robot skill learning from raw human videos. Compared with teleoperation or kinesthetic teaching in the laboratory, such learning method can flexibly utilize large-scale human videos available on the Internet, thereby improving the robot’s ability to generalize to various complex scenarios. Previous works on learning from videos usually have some issues, including requirements for robot actions, consistent viewpoints, similar layouts and low sample efficiency. To alleviate these issues, we propose a three-stage skill learning framework CIA. Temporal alignment is utilized to learn task priors through our proposed transformer-based model and self-supervised loss functions. A trajectory generation model is trained to learn the action priors. To further adapt to diverse scenarios, we propose a two-stage policy improvement method by initialization and interaction. An optimization method is introduced to ensure safe interaction and sample efficiency, where the optimization objective is guided by the learned task priors. The experimental results show that our CIA outperforms other state-of-the-art methods in task success rate and generalization to novel scenarios.

Given the rapid advancement of robotic technologies, robots will eventually enter our daily lives, performing complex long-horizon tasks. Complex long-horizon tasks, such as furniture assembly, typically contain dozens of subtasks with various scenes. Although complex, assembly is based on several reusable movements. Movement Primitive (MP) is a promising framework for learning reusable movements from demonstrations and adapting the learned movements to the test scenes. The critical step in employing MP methods is categorizing the unlabeled demonstrations into different MPs. However, current MP methods focus on individual MP learning using manually selected demonstrations, neglecting categorization. Manual categorization of demonstrations is easy to fall into suboptimal. If the demonstrations within the same category are too similar, the learned MP cannot be adapted to task scenes with various obstacles. Conversely, a significant distance between demonstrations leads to the MP’s failure in learning. To this end, we propose the following principle for MP categorization: balance the Learnability and Adaptability. Following this principle, we introduce an optimal transportation (OT)-based theoretical framework and a practical solution utilizing an auto-encoder network. We obtain the lower threshold of learnability by OT. Then we increase the adaptability of MP until it reaches the lower threshold of learnability. For complex long-horizon task learning, we propose a balanced MPs-based learning framework that contains four modules, termed BaMPs. BaMPs achieved success rates of 100% and 80%, respectively, in the 12-step and 20-step tasks.

Understanding the part composition and structure of 3D shapes is crucial for a wide range of 3D applications, including 3D part assembly and 3D assembly completion. Compared to 3D part assembly, 3D assembly completion is more complicated, which involves repairing broken or incomplete furniture that miss several parts with a toolkit. Given an incomplete assembly, 3D assembly completion seeks to identify its missing parts from multiple candidates, determine their poses, and produce complete assembly that is well-connected, structurally stable, and aesthetically pleasing. This task necessitates not only specialized knowledge of part composition but, more importantly, an awareness of physical constraints, i.e., connectivity, stability, and symmetry. Neglecting these constraints often results in assemblies that, although visually plausible, are impractical. To address this challenge, we propose PhysFiT, a physical-aware 3D shape understanding framework. This framework is built upon attention-based part relation modeling and incorporates connection modeling, simulation-free stability optimization and symmetric transformation consistency. We evaluate its efficacy on 3D part assembly and 3D assembly completion, a novel assembly task presented in this work. Extensive experiments demonstrate the effectiveness of PhysFiT in constructing geometrically sound and physically compliant assemblies.

Automatic assembly is a promising research topic in 3D computer vision and robotics. Existing works focus on generating assembly (e.g., IKEA furniture) from scratch with a set of parts, namely 3D part assembly. In practice, there are higher demands for the robot to take over and finish an incomplete assembly (e.g., a half-assembled IKEA furniture) with an off-the-shelf toolkit, especially in human-robot and multi-agent collaborations. Compared to 3D part assembly, it is more complicated in nature and remains unexplored yet. The robot must understand the incomplete structure, infer what parts are missing, single out the correct parts from the toolkit and finally, assemble them with appropriate poses to finish the incomplete assembly. Geometrically similar parts in the toolkit can interfere, and this problem will be exacerbated with more missing parts. To tackle this issue, we propose a novel task called 3D assembly completion. Given an incomplete assembly, it aims to find its missing parts from a toolkit and predict the 6-DoF poses to make the assembly complete. To this end, we propose FiT, a framework for Finishing the incomplete 3D assembly with Transformer. We employ the encoder to model the incomplete assembly into memories. Candidate parts interact with memories in a memory-query paradigm for final candidate classification and pose prediction. Bipartite part matching and symmetric transformation consistency are embedded to refine the completion. For reasonable evaluation and further reference, we design two standard toolkits of different difficulty, containing different compositions of candidate parts. We conduct extensive comparisons with several baseline methods and ablation studies, demonstrating the effectiveness of the proposed method.

The migrative functional equations provide a very powerful tool for constructing and characterizing new fuzzy logic connectives by convex combination, and have particularly important applications in image processing. So far, the migrativity between conjunctive logic connectives has been extensively studied, the obtained results do not, however, work well for triangular conorms. This paper is devoted to an in-depth investigation on three types of migrativity for continuous triangular conorms S with respect to N -ordinal sum implications I , which have distinctive features from ordinary ordinal sum implications. We will provide first detailed characterizations on the ( α , I ) -migrativity of S for each case according to the position relation of α in the range of N , by giving the corresponding ordinal sum decompositions of the t-conorm and implication solutions. Then the ( α , I ) -migrativity is extended to internal and global cases, and their characterizations are given under some additional constraints.

Motivated by several problems on constructions, algebraic properties, in particular the law of importation (LI) and the flexible ordering property (FOP), and applications in approximate reasoning, the present article defines a new class of fuzzy implications, called$(\theta,t)$-generated implications, by a pair of multiplicative and additive generators oft-norms and the usual addition instead of the multiplication or division usually used in the literature. The relations to other generator generated implications are clarified, the intersections with$(S,N)$-implications are discussed in detail and the intersections with$R$-implications are proved to be conjugates of the Łukasiewicz implication. Thet-norm solutions to the (LI) equation for$(\theta,t)$-generated implications in several cases are characterized in terms of generated t-norms, and two special subclasses of$(\theta,t)$-generated implications are characterized through (LI) and (FOP) on the other hand. A particularly interesting corollary shows that a binary operation on [0,1] enjoying (OP) satisfies (LI) with the Łukasiewiczt-norm if and only if it is the Łukasiewicz implication. These results will partially solve or enrich the related open problems pending for fuzzy implications. As applications of$(\theta,t)$-generated implications, two fuzzy reasoning methods, called triple I method and parallel hierarchical triple I method, based on (LI) and (FOP) are proposed in order to set a logic foundation for Zadeh’s compositional rule of inference (CRI) and to remedy the dependency of Jayaram’s hierarchical CRI on the order of inputs, respectively.

The present article proposes two construction ways to study the general forms of ordinal sums of fuzzy implications with the intent of unifying the ordinal sums existing in the literature. The first ordinal sum construction way, which we call “Implication Complementing,” is to study how to complement a specific fuzzy implication to the linear transformations of given fuzzy implications defined on respective disjoint subsquares whose principal diagonals are segments of the principal diagonal of the unit square, in order that the resulting ordinal sum on the unit square is a fuzzy implication. The second way, which we call “Implication Reconstructing,” is to study how to reconstruct an initial fuzzy implication through replacing its some values on given rectangular regions of the unit square with the linear transformations of respective given fuzzy implications such that the redefined function is a new fuzzy implication. In both ways, necessary and sufficient conditions for the final reconstructed functions to be fuzzy implications are given and several new constructions of ordinal sums of fuzzy implications are obtained, which would generalize the existing ordinal sums from several aspects. In particular, by adopting the idea behind the second way, the generalized ordinal sums of fuzzy implications are proposed, in which the regions where the linear transformations of given fuzzy implications are defined are neither necessarily subsquares nor necessarily along the principal or minor diagonal of the unit square.

In this article, we define the$k$-generated implications by continuous multiplicative generators$k$of$t$-norms, along with the research lines of Yager's$f$-generated implications by continuous additive generators$f$of$t$-norms and$g$-generated implications by continuous additive generators$g$of$t$-conorms and of Balasubramaniam's$h$-generated implications by continuous multiplicative generators$h$of$t$-conorms. This article is mainly motivated by many desirable properties of such generator generating implications for their flexible ordering property, unique determinations by induced natural negations, close relations to nilpotent$t$-norms,$T$-conditionality, the law of importation with more choices of$t$-norms$T$other than$T_{P}$, and good distributivity over$t$-norms and$t$-conorms. The relationships of$k$-generated implications to other well-known classes of fuzzy implications are clarified: they will unify$g$-generated implications with$g(1)< \infty$and$h$-generated implications with$h(1)>0$and hence, have intersections with$(S,N)$-implications; the only intersection with$R$-implications is the Goguen implication and they have no intersections with$f$-generated implications. The main results of this article are several characterizing theorems for (sometimes subclasses of)$k$-generated implications by means of aforementioned desirable properties from the respective perspectives, of which some partially solved or enriched related open problems in the literature. Finally, the superiorities of$k$-generated implications for modeling strict fuzzy preference relations and for constructing novel fuzzy reasoning methods are analyzed.

Liquid State Machine (LSM) is a biologically plausible neural network model for real-time computing on time-varying inputs, which is shown to be beneficial for perform computational tasks like pattern classification. The LSM uses spiking neurons with dynamic synapses to projects inputs into high-dimensional space, facilitating subsequent linear pattern recognition In this paper, we present two different methods to improve LSM in real-time pattern classification from perspectives of spatial integration and temporal integration. We develop the reservoir for LSM with a self-organizing network (SON) constructed by refining synaptic connectivity based on spike-time-dependent plasticity (STDP). Our study shows LSM with SON has better performance than LSM with random reservoir in spike train classification especially for small amount of templates due to spatial integration. For temporal integration, we show that increasing time constant of the linear filter, which transform spikes into smoothly changing states, is able to lead a better performance and is robust to the amount of templates.

Motif discovery in sequence data is fundamental to many biological problems such as antibody biomarker identification. Recent advances in instrumental techniques make it possible to generate thousands of protein sequences at once, which raises a big data issue for the existing motif finding algorithms: They either work only in a small scale of several hundred sequences or have to trade accuracy for efficiency. In this work, we demonstrate that by intelligently clustering sequences, it is possible to significantly improve the scalability of all the existing motif finding algorithms without losing accuracy at all. An anchor based sequence clustering algorithm (ASC) is thus proposed to divide a sequence dataset into multiple smaller clusters so that sequences sharing the same motif will be located into the same cluster. Then an existing motif finding algorithm can be applied to each individual cluster to generate motifs. In the end, the results from multiple clusters are merged together as final output. Experimental results show that our approach is generic and orders of magnitude faster than traditional motif finding algorithms. It can discover motifs from protein sequences in the scale that no existing algorithm can handle. In particular, ASC reduces the running time of a very popular motif finding algorithm, MEME, from weeks to a few minutes with even better accuracy.

Bacterial chemotaxis is one of the best studied signal transduction pathways. CheW is a scaffold protein that mediates the association of the chemoreceptors and the CheA kinase in a ternary signaling complex. The effects of replacing conserved Arg62 of CheW with other residues suggested that the scaffold protein plays a more complex role than simply binding its partner proteins. Although R62A CheW had essentially the same affinity for chemoreceptors and CheA, cells expressing the mutant protein are impaired in chemotaxis. Using a combination of molecular dynamics simulations (MD), NMR spectroscopy, and circular dichroism (CD), we addressed the role of Arg62. Here we show that Arg62 forms a salt bridge with another highly conserved residue, Glu38. Although this interaction is unimportant for overall protein stability, it is essential to maintain the correct alignment of the chemoreceptor and kinase binding sites of CheW. Computational and experimental data suggest that the role of the salt bridge in maintaining the alignment of the two partner binding sites is fundamental to the function of the signaling complex but not to its assembly. We conclude that a key feature of CheW is to maintain the specific geometry between the two interaction sites required for its function as a scaffold.

In this paper, we propose a hierarchical approach to solving sensor planning for the global localization of a mobile robot. Our system consists of two subsystems: a lower layer and a higher layer. The lower layer uses a particle filter to evaluate the posterior probability of the localization. When the particles converge into clusters, the higher layer starts particle clustering and sensor planning to generate an optimal sensing action sequence for the localization. The higher layer uses a Bayesian network for probabilistic inference. The sensor planning takes into account both localization belief and sensing cost. We conducted simulations and actual robot experiments to validate our proposed approach.

BACKGROUND: Since proteins perform their functions by interacting with one another and with other biomolecules, reconstructing a map of the protein-protein interactions of a cell, experimentally or computationally, is an important first step toward understanding cellular function and machinery of a proteome. Solely derived from the Gene Ontology (GO), we have defined an effective method of reconstructing a yeast protein interaction network by measuring relative specificity similarity (RSS) between two GO terms. DESCRIPTION: Based on the RSS method, here, we introduce a predicted Saccharomyces protein-protein interaction database called SPIDer. It houses a gold standard positive dataset (GSP) with high confidence level that covered 79.2% of the high-quality interaction dataset. Our predicted protein-protein interaction network reconstructed from the GSPs consists of 92,257 interactions among 3600 proteins, and forms 23 connected components. It also provides general links to connect predicted protein-protein interactions with three other databases, DIP, BIND and MIPS. An Internet-based interface provides users with fast and convenient access to protein-protein interactions based on various search features (searching by protein information, GO term information or sequence similarity). In addition, the RSS value of two GO terms in the same ontology, and the inter-member interactions in a list of proteins of interest or in a protein complex could be retrieved. Furthermore, the database presents a user-friendly graphical interface which is created dynamically for visualizing an interaction sub-network. The database is accessible at http://cmb.bnu.edu.cn/SPIDer/index.html. CONCLUSION: SPIDer is a public database server for protein-protein interactions based on the yeast genome. It provides a variety of search options and graphical visualization of an interaction network. In particular, it will be very useful for the study of inter-member interactions among a list of proteins, especially the protein complex. In addition, based on the predicted interaction dataset, researchers could analyze the whole interaction network and associate the network topology with gene/protein properties based on a global or local topology view.

In this paper we propose a hierarchical approach to solving sensor planning for the global localization of a mobile robot. Our system consists of two subsystems: a lower and a higher layer. The lower layer uses a particle filter to evaluate the posterior probability of the localization. When the particles converge into clusters, the higher layer starts particle clustering and sensor planning to generate an optimal sensing action sequence for the localization. The higher layer uses a Bayesian network for the probabilistic inference. The sensor planning takes into account both localization belief and sensing cost. We conducted simulations and actual robot experiments to validate our proposed approach.

We propose a novel method to solve a kidnapped robot problem. A mobile robot plans its sensor actions to localize itself using Bayesian network inference. The system differs from traditional methods such as the simple Bayesian decision or top-down action selection based on a decision tree. In contrast, we represent the contextual relation between the local sensing results and beliefs about the global localization using Bayesian networks. Inference of the Bayesian network allows us to classify ambiguous positions of the mobile robot when the local sensing evidences are obtained. By taking into account the trade-off between the global localization belief degree and local sensing cost, we define an integrated utility function to decide the local sensing range, and obtain an optimal sensing plan and optimal Bayesian network structure based on this function. We conducted simulation and real robot experiments to validate our planning concept.