Mingyang Sun

Research scientist / Engineer

I’m Mingyang Sun, a research scientist and engineer at Luma AI working on robot learning and embodied intelligence. I’m interested in how machines connect perception, language, and action to learn from and interact with the physical world.

My current research spans vision-language-action models, diffusion policies, and dexterous manipulation. I work across model post-training, reinforcement learning, and simulation to bring learned behavior closer to real-world deployment. My background in design and spatial systems shapes how I approach this work: connecting data, models, and interaction through hands-on experimentation.

Selected projects (06)

All projects (08)

Urban Fusion

A Multi-Constraint Site Selection Framework with LLM-Driven Spatial Reasoning

  1. Urban FusionAgents / spatial reasoning
  2. Date SyncMultimodal AI
  3. 404 Housing Not FoundData visualization
  4. RecitygenGenerative design
  5. Urban EchoesVisual similarity
  6. Cognitive LandscapesVR / EEG

Areas of interest (05)

  1. Embodied AI & Robot Learning

    Vision-language-action models, visuomotor learning, and robotic manipulation.

  2. Post-Training & Reinforcement Learning

    Adapting policies through demonstrations, rewards, and interaction with their environment.

  3. World Models & Spatial Intelligence

    Generative video and 3D models for understanding and predicting physical environments.

  4. Simulation & Sim-to-Real

    Physics fidelity, synthetic data, and transferring learning from simulation to the physical world.

  5. Evaluation & Scalable ML Systems

    Rigorous benchmarks, failure analysis, and reproducible infrastructure for research.

Contact

myangsun@outlook.comLinkedIn