2026 — Present
Nanyang Technological University
PhD Student
Advised by Prof. Jianfei Yang. Research direction: multimodal learning, MLLM Agents, and embodied intelligence.
PhD Student · Multimodal AI · Embodied Intelligence
Bohan Hou 侯博涵
I build multimodal foundation models and agents that connect perception, reasoning, world modeling, and action.
Nanyang Technological UniversityPhD Student
Alibaba DAMO AcademyResearch Intern
Recent updates
InterLV-Search, ECL-Bench, and RynnVLA-002 accepted to NeurIPS 2026.
Our World Model for Robot Learning survey accepted to IJRR.
01 / Profile
About
I am a PhD student at Nanyang Technological University (NTU), advised by Prof. Jianfei Yang. My research focuses on multimodal foundation models and agents for embodied intelligence, particularly embodied reasoning, vision-language-action models, world models, and robot learning.
I received my B.Eng. in Computer Science from Taishan College at Shandong University, where I worked with Prof. Xuemeng Song, and I have been a research assistant at the Intelligent Media Research Center (iLearn), led by Prof. Liqiang Nie. I am also a research intern at Alibaba DAMO Academy, developing open embodied models and systems.
02 / Background
Education
Academic training and degree programs.
Sep. 2022 — Jun. 2026
Shandong University
B.Eng. in Computer Science and Technology
Taishan College, First-Honor Program. GPA: 88.5 / 100; Rank: 5 / 22.
03 / Appointments
Experience
Research appointments and internships.
Sep. 2022 — Present
iLearn Lab, Shandong University
Research Assistant · Multimodal Large Language Models and Information Retrieval
Supervisors: Prof. Xuemeng Song and Prof. Liqiang Nie.
04 / Research
Publications
* Equal or core contribution, as specified by each paper.
Showing 15 of 15 works
InterLV-Search: Benchmarking Interleaved Multimodal Agentic Search
* Bohan Hou and Jiuning Gu contributed equally.
A benchmark for interleaved language-vision agentic search with active visual evidence seeking and open-web multimodal search.
AgentInformation RetrievalMLLMs
Benchmarking Fine-Grained Spatio-Temporal Awareness in Embodied Brain Models
* Minghao Zhu, Zhikai Wang, Ronghao Dang, and Bohan Hou contributed equally.
ECL-Bench evaluates fine-grained spatial and temporal awareness in embodied brain models.
MLLMsEmbodied AI
RynnVLA-002: A Unified Vision-Language-Action and World Model
A unified autoregressive VLA and world model in which action generation and future-frame prediction improve one another for simulated and real-world robot manipulation.
World ModelEmbodied AI
World Model for Robot Learning: A Comprehensive Survey
A robotics-centered survey of world models for policy learning, planning, learned simulation, evaluation, and video generation.
World ModelEmbodied AI
MARS Policy: Multimodality Only When It Matters
A modality-adaptive robot policy that introduces stochasticity when action diversity is useful, while retaining efficient deterministic behavior during single-modal phases.
Embodied AI
Thinking Beyond Videos: Unifying Video Reasoning and Deep Research for Open-World Video Agents
VideoRover unifies active video reasoning and deep research through iterative video cropping, multimodal search, and webpage browsing, with curated training trajectories and VideoRover-Bench.
AgentInformation RetrievalMLLMs
Data Pyramid for Embodied Manipulation
A data-centric taxonomy that organizes embodied manipulation data into five complementary layers and connects their collection, alignment, and mixing strategies to embodied foundation models.
Embodied AI
RynnBrain 1.1: Towards More Capable and Generalizable Embodied Foundation Model
An embodied foundation-model family spanning 2B, 9B, and 122B-A10B scales, with stronger cognition and localization, native 3D grounding, contact-point prediction, and cross-embodiment VLA capabilities.
MLLMsEmbodied AI
RynnValue: Scaling Robotic Value Foundation Models with Temporal Distance
A general-purpose robotic value foundation model that learns temporal distance from 7,000+ hours of heterogeneous data, turning time-to-completion into signals for progress estimation, failure detection, and VLA reinforcement learning.
Embodied AI
RynnBrain: Open Embodied Foundation Models
An open embodied foundation model family for egocentric understanding, spatiotemporal localization, physical reasoning, and planning.
MLLMsEmbodied AI
FiRE: Enhancing MLLMs with Fine-Grained Context Learning for Complex Image Retrieval
Fine-grained context learning for complex image retrieval with multimodal large language models.
Information RetrievalMLLMs
Visual Anchor Point for Multimodal Document Retrieval
Winner of the WWW’25 Multimodal Document Retrieval Challenge.
Information Retrieval
Pseudo-triplet Guided Few-shot Composed Image Retrieval
A pseudo-triplet guided scheme for improving few-shot composed image retrieval.
Information Retrieval
ImgEdit: A Unified Image Editing Dataset and Benchmark
A unified image editing dataset, benchmark, and model for high-quality single-turn and multi-turn image editing.
MLLMs
A Comprehensive Survey on Composed Image Retrieval
A comprehensive survey of composed image retrieval, covering supervised, zero-shot, and related multimodal retrieval settings.
Information Retrieval
No papers match these filters.
05 / Open Source
Selected Project
Embodied-AI-Guide
Open sourceA popular Chinese knowledge base and resource index for embodied intelligence, organized in an encyclopedia-style format. It covers robotics foundations, robot learning, simulators, benchmarks, datasets, multimodal models, VLA models, navigation, and related embodied-AI research directions.
06 / More
Honors & Contact
Selected Honors
- Champion, WWW’25 Multimodal Document Retrieval Challenge.
- Shandong University Outstanding Student Scholarship, 2023–2025.
- Shandong University Academic Innovation Scholarship, 2025.
- Second Prize, National Undergraduate Mathematics Competition, 2024.