Zihao Wang

Kaichen He, Zihao Wang, Muyao Li et al.

The paradigm of agentic AI is shifting from engineered complex workflows to post-training native models. However, existing agents are typically confined to static, predefined action spaces--such as exclusively using APIs, GUI events, or robotic commands. This rigidity limits their adaptability in dynamic environments where the optimal granularity of interaction varies contextually. To bridge this gap, we propose CrossAgent, a unified agentic model that masters heterogeneous action spaces and autonomously selects the most effective interface for each step of a trajectory. We introduce a comprehensive training pipeline that integrates cold-start supervised fine-tuning with a Multi-Turn Group Relative Policy Optimization (GRPO) algorithm. This approach enables the agent to learn adaptive action switching--balancing high-level efficiency with low-level precision--without human-specified rules. Extensive experiments on over 800 tasks in the open-world Minecraft environment demonstrate that CrossAgent achieves state-of-the-art performance. By dynamically leveraging the strengths of diverse action spaces, our model significantly outperforms fixed-action baselines, exhibiting superior generalization and efficiency in long-horizon reasoning. All code and models are available at https://github.com/CraftJarvis/OpenHA

Zihao Wang

The proposed BSDE-based diffusion model represents a novel approach to diffusion modeling, which extends the application of stochastic differential equations (SDEs) in machine learning. Unlike traditional SDE-based diffusion models, our model can determine the initial conditions necessary to reach a desired terminal distribution by adapting an existing score function. We demonstrate the theoretical guarantees of the model, the benefits of using Lipschitz networks for score matching, and its potential applications in various areas such as diffusion inversion, conditional diffusion, and uncertainty quantification. Our work represents a contribution to the field of score-based generative learning and offers a promising direction for solving real-world problems.

Jianqin Gao, Tianqi Wang, Yu Zhang et al.

The widespread use of mobile devices has created new challenges for vision systems in safety monitoring, workplace productivity assessment, and attention management. Detecting whether a person is using a phone requires not only object recognition but also an understanding of behavioral context, which involves reasoning about the relationship between faces, hands, and devices under diverse conditions. Existing generic benchmarks do not fully capture such fine-grained human--device interactions. To address this gap, we introduce the FPI-Det, containing 22{,}879 images with synchronized annotations for faces and phones across workplace, education, transportation, and public scenarios. The dataset features extreme scale variation, frequent occlusions, and varied capture conditions. We evaluate representative YOLO and DETR detectors, providing baseline results and an analysis of performance across object sizes, occlusion levels, and environments. Source code and dataset is available at https://github.com/KvCgRv/FPI-Det.

Zihao Wang, Anji Liu, Haowei Lin et al. · pku

We explore how iterative revising a chain of thoughts with the help of information retrieval significantly improves large language models' reasoning and generation ability in long-horizon generation tasks, while hugely mitigating hallucination. In particular, the proposed method -- *retrieval-augmented thoughts* (RAT) -- revises each thought step one by one with retrieved information relevant to the task query, the current and the past thought steps, after the initial zero-shot CoT is generated. Applying RAT to GPT-3.5, GPT-4, and CodeLLaMA-7b substantially improves their performances on various long-horizon generation tasks; on average of relatively increasing rating scores by 13.63% on code generation, 16.96% on mathematical reasoning, 19.2% on creative writing, and 42.78% on embodied task planning. The demo page can be found at https://craftjarvis.github.io/RAT

Shaofei Cai, Zihao Wang, Kewei Lian et al. · pku

Vision-language models (VLMs) have excelled in multimodal tasks, but adapting them to embodied decision-making in open-world environments presents challenges. One critical issue is bridging the gap between discrete entities in low-level observations and the abstract concepts required for effective planning. A common solution is building hierarchical agents, where VLMs serve as high-level reasoners that break down tasks into executable sub-tasks, typically specified using language. However, language suffers from the inability to communicate detailed spatial information. We propose visual-temporal context prompting, a novel communication protocol between VLMs and policy models. This protocol leverages object segmentation from past observations to guide policy-environment interactions. Using this approach, we train ROCKET-1, a low-level policy that predicts actions based on concatenated visual observations and segmentation masks, supported by real-time object tracking from SAM-2. Our method unlocks the potential of VLMs, enabling them to tackle complex tasks that demand spatial reasoning. Experiments in Minecraft show that our approach enables agents to achieve previously unattainable tasks, with a $\mathbf{76}\%$ absolute improvement in open-world interaction performance. Codes and demos are now available on the project page: https://craftjarvis.github.io/ROCKET-1.

Guangyu Zhao, Kewei Lian, Haowei Lin et al. · pku

A glowing body of work has emerged focusing on instruction-following policies for open-world agents, aiming to better align the agent's behavior with human intentions. However, the performance of these policies is highly susceptible to the initial prompt, which leads to extra efforts in selecting the best instructions. We propose a framework named Preference Goal Tuning (PGT). PGT allows an instruction following policy to interact with the environment to collect several trajectories, which will be categorized into positive and negative samples based on preference. Then we use preference learning to fine-tune the initial goal latent representation with the categorized trajectories while keeping the policy backbone frozen. The experiment result shows that with minimal data and training, PGT achieves an average relative improvement of 72.0% and 81.6% over 17 tasks in 2 different foundation policies respectively, and outperforms the best human-selected instructions. Moreover, PGT surpasses full fine-tuning in the out-of-distribution (OOD) task-execution environments by 13.4%, indicating that our approach retains strong generalization capabilities. Since our approach stores a single latent representation for each task independently, it can be viewed as an efficient method for continual learning, without the risk of catastrophic forgetting or task interference. In short, PGT enhances the performance of agents across nearly all tasks in the Minecraft Skillforge benchmark and demonstrates robustness to the execution environment.

Zihao Wang, Clair Vandersteen, Thomas Demarcy et al.

Signed distance map (SDM) is a common representation of surfaces in medical image analysis and machine learning. The computational complexity of SDM for 3D parametric shapes is often a bottleneck in many applications, thus limiting their interest. In this paper, we propose a learning based SDM generation neural network which is demonstrated on a tridimensional cochlea shape model parameterized by 4 shape parameters. The proposed SDM Neural Network generates a cochlea signed distance map depending on four input parameters and we show that the deep learning approach leads to a 60 fold improvement in the time of computation compared to more classical SDM generation methods. Therefore, the proposed approach achieves a good trade-off between accuracy and efficiency.