MLL Lab
MLL Lab (Multimodal Learning and Language Lab) is an academic research lab directed by Northwestern Assistant Professor Manling Li. It develops intelligent language and multimodal models that reason, plan, and interact with the physical world, with work spanning language, vision, robotics, embodied agents, and foundation-model reasoning.
At a Glance
- Academic and industrial AI research teams
- Developers building LLM and VLM agents
- Robotics and embodied-AI researchers
- Foundation-model and multimodal-learning researchers
- +1 more
AI Tools by MLL Lab
(1)
RAGEN
Open Source RL Agent Trainer
Discussions
No discussions yet
Be the first to start a discussion about MLL Lab
Latest News
MLL Lab co-organized the KnowFM workshop at ACL 2026.
MLL Lab organized the FAGEN: Failure Modes in Agentic AI workshop at ICML 2026.
RAGEN-2 received the Best Paper Award (Top 1%) at the CVPR 2026 MMRAgI Workshop.
RAGEN-2 was accepted to ICML 2026 as an Oral Presentation (Top 0.7%).
Products & Services
Open-source framework for training LLM reasoning agents with multi-turn reinforcement learning in stochastic environments; the project uses an MDP formulation and reasoning/interaction optimization, with StarPO and reasoning-collapse diagnostics in the current repository.
Research project studying reasoning collapse in agentic reinforcement learning, including template and low-entropy failure modes and methods to preserve diverse, input-grounded reasoning.
World-model reinforcement-learning framework for multi-turn VLM agents; its TRICO method focuses on critical tokens and cross-turn credit assignment.
Unified evaluation interface for embodied decision-making that supports diverse tasks and LLM modules and provides fine-grained error metrics such as hallucination and affordance errors.
Market Position
MLL Lab is positioned as a university research lab focused on the intersection of multimodal foundation models, agentic reinforcement learning, spatial intelligence, and embodied AI. Its differentiators are open research artifacts (code, datasets, benchmarks, and documentation), multi-turn agent training, and explicit study of reasoning failure modes and physical-world grounding rather than a commercial SaaS product.
Leadership
Founders
Manling Li
Assistant Professor of Computer Science at Northwestern University, affiliated with Northwestern's Center for Robotics and Biosystems and Cognitive Science, and an Amazon Scholar. Previously a Stanford postdoctoral researcher working with Jiajun Wu and Fei-Fei Li; earned a 2023 PhD from UIUC advised by Heng Ji and worked closely with Shih-Fu Chang, Kyunghyun Cho, and Jiawei Han.
Executive Team
Manling Li
Principal Investigator and Director
Northwestern Assistant Professor of Computer Science and Amazon Scholar; former Stanford postdoc and UIUC PhD, researching multimodal and embodied reasoning, agent training, planning, and trustworthy AI.
Founding Story
MLL Lab officially started in October 2024 under Manling Li after she joined Northwestern University. Its initial vision is to move foundation models from passive observers toward active, embodied agents by combining language, vision, robotics, reasoning, planning, and grounded interaction with the physical world.
Business Model
Revenue Model
Academic research lab; its public materials describe research, open-source software, datasets, benchmarks, documentation, publications, and student recruitment rather than a commercial revenue model.
Target Markets
- Academic and industrial AI research teams
- Developers building LLM and VLM agents
- Robotics and embodied-AI researchers
- Foundation-model and multimodal-learning researchers
- Researchers studying reinforcement learning, spatial intelligence, and world models
- Training LLM agents that reason and act over multiple turns
- Training VLM agents for visual and embodied tasks
- Robotics and physical-world interaction
- Spatial intelligence and 3D scene/view planning
- Benchmarking embodied cognition and agent reliability
- Long-video temporal retrieval and understanding