Han Lu
Researcher at 
Meituan LongCat Team
Previously CS PhD Student at 
About
Han Lu’s ROLL Flash uses asynchrony to accelerate agentic training and reinforcement learning from variable-length rewards. They are a researcher at the Meituan LongCat Team and a PhD graduate of Shanghai Jiao Tong University. Lu’s research focus includes RL, LLM pretraining, and CV, exemplified by the development of ActiveAD for planning-oriented active learning in autonomous driving. They also co-authored Detailed 2D-3D Joint Representation for human-object interaction modeling and the ROCO framework for evaluating combinatorial optimization solvers. Lu previously held research internships at Alibaba, ByteDance, Bilibili, and Huawei Noah's Ark Lab, where they worked on scalable MoE architectures and LLM reasoning policies. They completed both their CS BEng and PhD degrees at SJTU as a member of the ACM and Wu Wenjun AI honors classes.
Experience
Coach
2021 – Present
Jiangsu Province Informatics Olympiad Winter Camp
2021 – Present
Teaching Assistant
2021 – 2026
TA for Programming & Data Structures.
2021 – 2026
CS PhD Student
2021 – 2026
Advised by Prof. Junchi Yan; part of the Wu Wenjun AI Honors Class.
2021 – 2026
Vice Class Advisor
2021 – 2023
Served as vice class advisor for SJTU ACM Class, Class of 2023.
2021 – 2023
CS BEng Student
2017 – 2021
Part of the Zhiyuan ACM Honors Class, ranked in the top 5%.
2017 – 2021
Research Intern
Jun 2025 – Oct 2025
ALIBABA
Worked in Future Life Lab (T-STAR) on asynchronous RL framework design for LLM reasoning and off-policy RL for RLVR acceleration.
Jun 2025 – Oct 2025
Research Intern
Mar 2025 – Jun 2025
Worked in Douyin Recommendation (Ranking), exploring scalable FFN, MoE, and Transformer architectures for long-sequence user modeling in ranking models.
Mar 2025 – Jun 2025
Research Intern
Jun 2024 – Feb 2025
BILIBILI
Worked in the Foundation Model Group on LLM continual pretraining, MoE model deployment on GPU/NPU, and AI-assisted dubbing with LLMs.
Jun 2024 – Feb 2025
Papers12
Detailed 2D-3D Joint Representation for Human-Object Interaction
Human-Object Interaction (HOI) detection lies at the core of action understanding. Besides 2D information such as human/object appearance and locations, 3D pose is also usually utilized in HOI learning since its view-independence. However, rough 3D body joints just carry sparse body information and are not sufficient to understand complex interactions. Thus, we need detailed 3D body shape to go further. Meanwhile, the interacted object in 3D is also not fully studied in HOI learning. In light of these, we propose a detailed 2D-3D joint representation learning method. First, we utilize the single-view human body capture method to obtain detailed 3D body, face and hand shapes. Next, we estimate the 3D object location and size with reference to the 2D human-object spatial configuration and object category priors. Finally, a joint learning framework and cross-modal consistency tasks are proposed to learn the joint HOI representation. To better evaluate the 2D ambiguity processing capacity of models, we propose a new benchmark named Ambiguous-HOI consisting of hard ambiguous images. Extensive experiments in large-scale HOI benchmark and Ambiguous-HOI show impressive effectiveness of our method. Code and data are available at this https URL.
Learning-Aided Heuristics Design for Storage System
Computer systems such as storage systems normally require transparent white-box algorithms that are interpretable for human experts. In this work, we propose a learning-aided heuristic design method, which automatically generates human-readable strategies from Deep Reinforcement Learning (DRL) agents. This method benefits from the power of deep learning but avoids the shortcoming of its black-box property. Besides the white-box advantage, experiments in our storage productions resource allocation scenario also show that this solution outperforms the systems default settings and the elaborately handcrafted strategy by human experts.
Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimization
The reasoning pattern of Large language models (LLMs) remains opaque, and reinforcement learning (RL) typically applies uniform credit across an entire generation, blurring the distinction between pivotal and routine steps. This work positions attention as a privileged substrate that renders the internal logic of LLMs legible, not merely as a byproduct of computation, but as a mechanistic blueprint of reasoning itself. We first distinguish attention heads between locally and globally focused information processing and reveal that locally focused heads produce a sawtooth pattern near the diagonal indicating phrasal chunks, while globally focused heads expose tokens that exert broad downstream influence over future tokens. We formalize these with two metrics: 1) Windowed Average Attention Distance, which measures the extent of backward attention within a clipped window; 2) Future Attention Influence, which quantifies a token's global importance as the average attention it receives from subsequent tokens. Taken together, these signals reveal a recurring preplan-and-anchor mechanism, where the model first performs a long-range contextual reference to generate an introductory token, which is immediately followed by or coincides with a semantic anchor token that organizes subsequent reasoning. Leveraging these insights, we introduce three novel RL strategies that dynamically perform targeted credit assignment to critical nodes (preplan tokens, anchor tokens, and their temporal coupling) and show consistent performance gains across various reasoning tasks. By aligning optimization with the model's intrinsic reasoning rhythm, we aim to transform opaque optimization into an actionable structure-aware process, hoping to offer a potential step toward more transparent and effective optimization of LLM reasoning.
Part II: ROLL Flash - Accelerating RLVR and Agentic Training with Asynchrony
Synchronous Reinforcement Learning (RL) post-training has emerged as a crucial step for enhancing Large Language Models (LLMs) with diverse capabilities. However, many systems designed to accelerate RL post-training still suffer from low resource utilization and limited scalability. We present ROLL Flash, a system that extends ROLL with native support for asynchronous RL post-training. ROLL Flash is built upon two core design principles: fine-grained parallelism and rollout-train decoupling. Guided by these principles, ROLL Flash provides flexible programming interfaces that enable a fully asynchronous training architecture and support efficient rollout mechanisms, including queue scheduling and environment-level asynchronous execution. Through comprehensive theoretical analysis and extensive experiments, we demonstrate that ROLL Flash significantly improves resource utilization and scalability over synchronous RL post-training. ROLL Flash achieves up to 2.24x speedup on RLVR tasks and 2.72x on agentic tasks, using the same GPU budget as synchronous baselines. Furthermore, we implement several popular off-policy algorithms and verify that asynchronous training can achieve performance on par with synchronous training.
Asymmetric Proximal Policy Optimization: mini-critics boost LLM reasoning
Most recent RL for LLMs (RL4LLM) methods avoid explicit critics, replacing them with average advantage baselines. This shift is largely pragmatic: conventional value functions are computationally expensive to train at LLM scale and often fail under sparse rewards and long reasoning horizons. We revisit this bottleneck from an architectural perspective and introduce Asymmetric Proximal Policy Optimization (AsyPPO), a simple and scalable framework that restores the critics role while remaining efficient in large-model settings. AsyPPO employs a set of lightweight mini-critics, each trained on disjoint prompt shards. This design encourages diversity while preserving calibration, reducing value-estimation bias. Beyond robust estimation, AsyPPO leverages inter-critic uncertainty to refine the policy update: (i) masking advantages in states where critics agree and gradients add little learning signal, and (ii) filtering high-divergence states from entropy regularization, suppressing spurious exploration. After training on open-source data with only 5,000 samples, AsyPPO consistently improves learning stability and performance across multiple benchmarks over strong baselines, such as GRPO, achieving performance gains of more than six percent on Qwen3-4b-Base and about three percent on Qwen3-8b-Base and Qwen3-14b-Base over classic PPO, without additional tricks. These results highlight the importance of architectural innovations for scalable, efficient algorithms.
ActiveAD: Planning-Oriented Active Learning for End-to-End Autonomous Driving
End-to-end differentiable learning for autonomous driving (AD) has recently become a prominent paradigm. One main bottleneck lies in its voracious appetite for high-quality labeled data e.g. 3D bounding boxes and semantic segmentation, which are notoriously expensive to manually annotate. The difficulty is further pronounced due to the prominent fact that the behaviors within samples in AD often suffer from long tailed distribution. In other words, a large part of collected data can be trivial (e.g. simply driving forward in a straight road) and only a few cases are safety-critical. In this paper, we explore a practically important yet under-explored problem about how to achieve sample and label efficiency for end-to-end AD. Specifically, we design a planning-oriented active learning method which progressively annotates part of collected raw data according to the proposed diversity and usefulness criteria for planning routes. Empirically, we show that our planning-oriented approach could outperform general active learning methods by a large margin. Notably, our method achieves comparable performance with state-of-the-art end-to-end AD methods - by using only 30% nuScenes data. We hope our work could inspire future works to explore end-to-end AD from a data-centric perspective in addition to methodology efforts.
Continual Predictive Learning from Videos
Predictive learning ideally builds the world model of physical processes in one or more given environments. Typical setups assume that we can collect data from all environments at all times. In practice, however, different prediction tasks may arrive sequentially so that the environments may change persistently throughout the training procedure. Can we develop predictive learning algorithms that can deal with more realistic, non-stationary physical environments? In this paper, we study a new continual learning problem in the context of video prediction, and observe that most existing methods suffer from severe catastrophic forgetting in this setup. To tackle this problem, we propose the continual predictive learning (CPL) approach, which learns a mixture world model via predictive experience replay and performs test-time adaptation with non-parametric task inference. We construct two new benchmarks based on RoboNet and KTH, in which different tasks correspond to different physical robotic environments or human actions. Our approach is shown to effectively mitigate forgetting and remarkably outperform the na\"ive combinations of previous art in video prediction and continual learning.
Metrics
Citations per year
Coauthors
