Research on PKU ZhiClasshttps://zhi-class.ai/en/research/Recent content in Research on PKU ZhiClassHugo -- gohugo.ioenCopyright © Zhi Class 2022-2026Thu, 12 Feb 2026 00:00:00 +0000LDA-1B: Scaling Latent Dynamics Action Model via Universal Embodied Data Ingestionhttps://zhi-class.ai/en/research/2602.lda-1b/Thu, 12 Feb 2026 00:00:00 +0000https://zhi-class.ai/en/research/2602.lda-1b/Recent robot foundation models largely rely on large-scale behavior cloning, which imitates expert actions but discards transferable dynamics knowledge embedded in heterogeneous embodied data.NavSpace: How Navigation Agents Follow Spatial Intelligence Instructionshttps://zhi-class.ai/en/research/2601.navspace/Sat, 31 Jan 2026 00:00:00 +0000https://zhi-class.ai/en/research/2601.navspace/Instruction-following navigation is a key step toward embodied intelligence.Learning Physics-Grounded 4D Dynamics with Neural Gaussian Force Fieldshttps://zhi-class.ai/en/research/2601.learning-physics-grounded-4d-dynamics-with-neura/Mon, 26 Jan 2026 00:00:00 +0000https://zhi-class.ai/en/research/2601.learning-physics-grounded-4d-dynamics-with-neura/Predicting physical dynamics from visual data remains a fundamental challenge in AI, as it requires both accurate scene understanding and robust physics reasoning.MCPMark: A Benchmark for Stress-Testing Realistic and Comprehensive MCP Usehttps://zhi-class.ai/en/research/2601.mcpmark/Mon, 26 Jan 2026 00:00:00 +0000https://zhi-class.ai/en/research/2601.mcpmark/The MCP standardizes how LLMs interact with external systems, forming the foundation for general agents.Neural Force Field: Few shot learning of generalized physical reasoninghttps://zhi-class.ai/en/research/2601.neural-force-field/Mon, 26 Jan 2026 00:00:00 +0000https://zhi-class.ai/en/research/2601.neural-force-field/We present NFF, a modeling framework built on NODE that learns interpretable force field representations which can be efficiently integrated through an ODE solver to predict object trajectories.Generalized Threshold Optimization with Harmony Multi-Threshold Neurons for Accurate ANN-to-SNN Conversionhttps://zhi-class.ai/en/research/2601.generalized-threshold-optimization-with-harmony/Tue, 20 Jan 2026 00:00:00 +0000https://zhi-class.ai/en/research/2601.generalized-threshold-optimization-with-harmony/Spiking Neural Networks (SNNs) are a promising paradigm designed to emulate the brain’s energy efficient by incorporating the timing of spikes.Luminark: Training-free, Probabilistically-Certified Watermarking for General Vision Generative Modelshttps://zhi-class.ai/en/research/2601.luminark/Sat, 03 Jan 2026 00:00:00 +0000https://zhi-class.ai/en/research/2601.luminark/In this paper, we introduce \emph{Luminark｝, a training-free and probabilistically-certified watermarking method for general vision generative models.CorrectNav: Self-Correction Flywheel Empowers Vision-Language-Action Navigation Modelhttps://zhi-class.ai/en/research/2511.correctnav/Sat, 08 Nov 2025 00:00:00 +0000https://zhi-class.ai/en/research/2511.correctnav/Existing vision-and-language navigation models often deviate from the correct trajectory when executing instructions.ToolVQA: A Dataset for Multi-step Reasoning VQA with External Toolshttps://zhi-class.ai/en/research/2507.toolvqa/Mon, 14 Jul 2025 00:00:00 +0000https://zhi-class.ai/en/research/2507.toolvqa/SimLauncher: Launching Sample-Efficient Real-world Robotic Reinforcement Learning via Simulation Pre-traininghttps://zhi-class.ai/en/research/2507.simlauncher/Sun, 06 Jul 2025 00:00:00 +0000https://zhi-class.ai/en/research/2507.simlauncher/Autonomous learning of dexterous, long-horizon robotic skills has been a longstanding pursuit of embodied AI.Playing with Transformer at 30+ FPS via Next-Frame Diffusionhttps://zhi-class.ai/en/research/2506.nextframed/Sun, 01 Jun 2025 00:00:00 +0000https://zhi-class.ai/en/research/2506.nextframed/In this work, we present Next-Frame Diffusion (NFD), an autoregressive diffusion transformer that incorporates block-wise causal attention, enabling iterative sampling and efficient inference via parallel token generation within each frame.Apply Hierarchical-Chain-of-Generation to Complex Attributes Text-to-3D Generationhttps://zhi-class.ai/en/research/2505.gascol/Wed, 07 May 2025 00:00:00 +0000https://zhi-class.ai/en/research/2505.gascol/OmniPhysGS: 3D Constitutive Gaussians for General Physics-based Dynamics Generationhttps://zhi-class.ai/en/research/2504.omniphysgs/Thu, 24 Apr 2025 00:00:00 +0000https://zhi-class.ai/en/research/2504.omniphysgs/ChemAgent: Self-updating Memories in Large Language Models Improves Chemical Reasoninghttps://zhi-class.ai/en/research/2501.chemagent/Fri, 03 Jan 2025 00:00:00 +0000https://zhi-class.ai/en/research/2501.chemagent/We present ChemAgent, a novel framework designed to improve the performance of LLMs through a dynamic, self-updating library.ProgressGym: Alignment with a Millennium of Moral Progresshttps://zhi-class.ai/en/research/paper8/Tue, 10 Dec 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper8/Autonomous Character-Scene Interaction Synthesis from Text Instructionhttps://zhi-class.ai/en/research/paper10/Tue, 03 Dec 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper10/DexGraspNet 2.0: Learning Generative Dexterous Grasping in Large-scale Synthetic Cluttered Sceneshttps://zhi-class.ai/en/research/paper6/Wed, 06 Nov 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper6/Benchmarking Open-instruction 6-DoF Object Rearrangement and A VLM-based Approachhttps://zhi-class.ai/en/research/paper2/Mon, 14 Oct 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper2/Exploring Conditional Multi-Modal Prompts for Zero-shot HOI Detectionhttps://zhi-class.ai/en/research/paper4/Sun, 29 Sep 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper4/MacDiff: Unified Skeleton Modeling with Masked Conditional Diffusionhttps://zhi-class.ai/en/research/paper3/Sun, 29 Sep 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper3/Language Models Represent Beliefs of Self and Othershttps://zhi-class.ai/en/research/paper7/Sun, 21 Jul 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper7/Diff-BGM: A Diffusion Model for Video Background Music Generationhttps://zhi-class.ai/en/research/paper9/Mon, 17 Jun 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper9/Exploring the Potential of Large Foundation Models for Open-Vocabulary HOI Detectionhttps://zhi-class.ai/en/research/paper5/Mon, 17 Jun 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper5/Scaling up dynamic human-scene interaction modelinghttps://zhi-class.ai/en/research/paper1/Mon, 17 Jun 2024 00:00:00 +0000https://zhi-class.ai/en/research/paper1/