Jing Long

Tianle Zhang, Zhihao Yuan, Dafeng Chi et al.

Robotic autonomy in open-world environments is fundamentally limited by insufficient data diversity and poor cross-embodiment generalization. Existing robotic datasets are often limited in scale and task coverage, while relatively large differences across robot embodiments impede effective behavior knowledge transfer. To address these challenges, we propose JoyAI-RA, a vision-language-action (VLA) embodied foundation model tailored for generalizable robotic manipulation. JoyAI-RA presents a multi-source multi-level pretraining framework that integrates web data, large-scale egocentric human manipulation videos, simulation-generated trajectories, and real-robot data. Through training on heterogeneous multi-source data with explicit action-space unification, JoyAI-RA effectively bridges embodiment gaps, particularly between human manipulation and robotic control, thereby enhancing cross-embodiment behavior learning. JoyAI-RA outperforms state-of-the-art methods in both simulation and real-world benchmarks, especially on diverse tasks with generalization demands.

Chen Zhou, Haoran Sun, Hedan Yang et al.

Embodied intelligence is a key step towards Artificial General Intelligence (AGI), yet its development faces multiple challenges including data, frameworks, infrastructure, and evaluation systems. To address these issues, we have, for the first time in the industry, launched a cloud-based, thousand-GPU distributed training platform for embodied intelligence, built upon the widely adopted LeRobot framework, and have systematically overcome bottlenecks across the entire pipeline. At the data layer, we have restructured the data pipeline to optimize the flow of embodied training data. In terms of training, for the GR00T-N1.5 model, utilizing thousand-GPU clusters and data at the scale of hundreds of millions, the single-round training time has been reduced from 15 hours to just 22 minutes, achieving a 40-fold speedup. At the model layer, by combining variable-length FlashAttention and Data Packing, we have moved from sample redundancy to sequence integration, resulting in a 188% speed increase; π-0.5 attention optimization has accelerated training by 165%; and FP8 quantization has delivered a 140% speedup. On the infrastructure side, relying on high-performance storage, a 3.2T RDMA network, and a Ray-driven elastic AI data lake, we have achieved deep synergy among data, storage, communication, and computation. We have also built an end-to-end evaluation system, creating a closed loop from training to simulation to assessment. This framework has already been fully validated on thousand-GPU clusters, laying a crucial technical foundation for the development and application of next-generation autonomous intelligent robots, and is expected to accelerate the arrival of the era of human-machine integration.

Zhong Guan, Haoran Sun, Yongjian Guo et al.

In recent years, Vision-Language-Action (VLA) models have emerged as a crucial pathway towards general embodied intelligence, yet their training efficiency has become a key bottleneck. Although existing reinforcement learning (RL)-based training frameworks like RLinf can enhance model generalization, they still rely on synchronous execution, leading to severe resource underutilization and throughput limitations during environment interaction, policy generation (rollout), and model update phases (actor). To overcome this challenge, this paper, for the first time, proposes and implements a fully-asynchronous policy training framework encompassing the entire pipeline from environment interaction, rollout generation, to actor policy updates. Systematically drawing inspiration from asynchronous optimization ideas in large model RL, our framework designs a multi-level decoupled architecture. This includes asynchronous parallelization of environment interaction and trajectory collection, streaming execution for policy generation, and decoupled scheduling for training updates. We validated the effectiveness of our method across diverse VLA models and environments. On the LIBERO benchmark, the framework achieves throughput improvements of up to 59.25\% compared to existing synchronous strategies. When deeply optimizing separation strategies, throughput can be increased by as much as 126.67\%. We verified the effectiveness of each asynchronous component via ablation studies. Scaling law validation across 8 to 256 GPUs demonstrates our method's excellent scalability under most conditions.

Yucheng Guo, Yongjian Guo, Zhong Guan et al.

In video generation models, particularly world models, training large-scale video diffusion Transformers (such as DiT and MMDiT) poses significant computational challenges due to the extreme variance in sequence lengths within mixed-mode datasets. Existing bucket-based data loading strategies typically rely on "equal token length" constraints. This approach fails to account for the quadratic complexity of self-attention mechanisms, leading to severe load imbalance and underutilization of GPU resources. This paper proposes \textit{AdaptiveLoad}, an integrated optimization framework consisting of two core components: (1) A dual-constraint adaptive load balancing system, which eliminates long-sequence bottlenecks by simultaneously limiting memory consumption and computational load ($B \times S^p \le M_{\text{comp}}$); (2) A fused LayerNorm-Modulate CUDA kernel, which utilizes a D-tile coalesced reduction strategy to increase throughput and alleviate memory pressure. Experimental results on the Wan 2.1 world model demonstrate that our method reduces the computational imbalance rate from 39\% to 18.9\%, improves peak VRAM utilization efficiency by 22.7\%, and achieves an overall training throughput increase of 27.2\%.

Wen Huang, Haoran Sun, Yongjian Guo et al.

World Action Models (WAMs) are an emerging family of policies that tie robot action generation to future-observation modeling. In this work, we focus on the joint video--action modeling paradigm, where actions and imagined future observations are co-generated along a shared denoising or flow trajectory, so that perception, prediction, and control are coupled within one generative process. Existing WAMs typically realize this paradigm with a Mixture-of-Transformers (MoT), where video and action tokens interact through shared self-attention. This architecture can in principle assign a separate timestep $t_f$ to each predicted latent frame, yet current systems collapse this degree of freedom onto a single shared scalar $t$. Under the noise-as-masking view of Diffusion Forcing, this shared schedule imposes the unjustified prior that every predicted latent is equally reliable for action generation. We instead view the per-latent schedule as a \emph{learnable information-gating policy}: by changing a latent frame's noise level, the policy modulates the reliability of its Key/Value contribution to the action tokens. We propose \textbf{NoiseGate}, which combines independent per-latent timestep sampling during backbone training, a lightweight Gating Policy Network that emits per-latent time increments during denoising, and task-reward optimization that trains the schedule policy without hand-crafted shape priors. Built on a joint video--action MoT backbone, NoiseGate delivers consistent gains on diverse RoboTwin random-scene manipulation tasks.

Sirui Huang, Jing Long, Qian Li et al.

Sequential Recommendation (SR) predicts users next interactions by modeling the temporal order of their historical behaviors. Existing approaches, including traditional sequential models and generative recommenders, achieve strong performance but primarily rely on explicit interactions such as clicks or purchases while overlooking item exposures. This ignorance introduces selection bias, where exposed but unclicked items are misinterpreted as disinterest, and exposure bias, where unexposed items are treated as irrelevant. Effectively addressing these biases requires distinguishing between items that were "not exposed" and those that were "not of interest", which cannot be reliably inferred from correlations in historical data. Counterfactual reasoning provides a natural solution by estimating user preferences under hypothetical exposure, and Inverse Propensity Scoring (IPS) is a common tool for such estimation. However, conventional IPS methods are static and fail to capture the sequential dependencies and temporal dynamics of user behavior. To overcome these limitations, we propose Time aware Inverse Propensity Scoring (TIPS). Unlike traditional static IPS, TIPS effectively accounts for sequential dependencies and temporal dynamics, thereby capturing user preferences more accurately. Extensive experiments show that TIPS consistently enhances recommendation performance as a plug-in for various sequential recommenders. Our code will be publicly available upon acceptance.

Rui Li, Jing Long, Muge Qi et al.

To facilitate robust and trustworthy deployment of large language models (LLMs), it is essential to quantify the reliability of their generations through uncertainty estimation. While recent efforts have made significant advancements by leveraging the internal logic and linguistic features of LLMs to estimate uncertainty scores, our empirical analysis highlights the pitfalls of these methods to strike a harmonized estimation between indication, balance, and calibration, which hinders their broader capability for accurate uncertainty estimation. To address this challenge, we propose CUE (Corrector for Uncertainty Estimation): A straightforward yet effective method that employs a lightweight model trained on data aligned with the target LLM's performance to adjust uncertainty scores. Comprehensive experiments across diverse models and tasks demonstrate its effectiveness, which achieves consistent improvements of up to 60% over existing methods.

Haocheng Tang, Jing Long, Beihong Ji et al.

In this work, we introduce Auxiliary Discriminator Sequence Generative Adversarial Networks (ADSeqGAN), a novel approach for molecular generation in small-sample datasets. Traditional generative models often struggle with limited training data, particularly in drug discovery, where molecular datasets for specific therapeutic targets, such as nucleic acids binders and central nervous system (CNS) drugs, are scarce. ADSeqGAN addresses this challenge by integrating an auxiliary random forest classifier as an additional discriminator into the GAN framework, significantly improves molecular generation quality and class specificity. Our method incorporates pretrained generator and Wasserstein distance to enhance training stability and diversity. We evaluate ADSeqGAN across three representative cases. First, on nucleic acid- and protein-targeting molecules, ADSeqGAN shows superior capability in generating nucleic acid binders compared to baseline models. Second, through oversampling, it markedly improves CNS drug generation, achieving higher yields than traditional de novo models. Third, in cannabinoid receptor type 1 (CB1) ligand design, ADSeqGAN generates novel druglike molecules, with 32.8\% predicted actives surpassing hit rates of CB1-focused and general-purpose libraries when assessed by a target-specific LRIP-SF scoring function. Overall, ADSeqGAN offers a versatile framework for molecular design in data-scarce scenarios, with demonstrated applications in nucleic acid binders, CNS drugs, and CB1 ligands.

Jing Long, Tong Chen, Nguyen Quoc Viet Hung et al.

Next Point-of-Interest (POI) recommendation has become an indispensable functionality in Location-based Social Networks (LBSNs) due to its effectiveness in helping people decide the next POI to visit. However, accurate recommendation requires a vast amount of historical check-in data, thus threatening user privacy as the location-sensitive data needs to be handled by cloud servers. Although there have been several on-device frameworks for privacy-preserving POI recommendations, they are still resource-intensive when it comes to storage and computation, and show limited robustness to the high sparsity of user-POI interactions. On this basis, we propose a novel decentralized collaborative learning framework for POI recommendation (DCLR), which allows users to train their personalized models locally in a collaborative manner. DCLR significantly reduces the local models' dependence on the cloud for training, and can be used to expand arbitrary centralized recommendation models. To counteract the sparsity of on-device user data when learning each local model, we design two self-supervision signals to pretrain the POI representations on the server with geographical and categorical correlations of POIs. To facilitate collaborative learning, we innovatively propose to incorporate knowledge from either geographically or semantically similar users into each local model with attentive aggregation and mutual information maximization. The collaborative learning process makes use of communications between devices while requiring only minor engagement from the central server for identifying user groups, and is compatible with common privacy preservation mechanisms like differential privacy. We evaluate DCLR with two real-world datasets, where the results show that DCLR outperforms state-of-the-art on-device frameworks and yields competitive results compared with centralized counterparts.