Home

Collaborate

Open problems. Let’s work on them together.

I’m looking for collaborators (faculty, industry researchers and students) who care about memory in perception and action. If one of these is your problem too, write to me.

  1. Efficient memory for long-horizon VLA policies

    Robot episodes run for minutes, and the world changes while they run. I bring memory-bank design from video (MAMBA, STPN).

    Good fit if you work on:
    • Robot learning
    • VLA
    • Long context
    Discuss this: Efficient memory for long-horizon VLA policies
  2. Memory as architecture

    What belongs in weights, and what belongs in context? Parametric memory (lookup tables, memory layers) and episodic memory (memory banks, KV caches, recurrent state) are usually studied apart. I want to know how they should divide the work when the stream is long and the agent has to act.

    Good fit if you work on:
    • Efficient attention
    • KV cache
    • Memory layers
    Discuss this: Memory as architecture
  3. Persistent world models for shared spaces

    World models that keep a state of the scene across long rollouts, including the changes people make, not only the effects of the robot’s own actions.

    Good fit if you work on:
    • Video generation
    • World models
    • Human–robot interaction
    Discuss this: Persistent world models for shared spaces
  4. Human motion and sign-language generation

    Building on Neural Sign Actors (led by colleagues at Imperial College London): more expressive hands and faces, and evaluation with Deaf and ASL-fluent signers.

    Good fit if you work on:
    • 3D humans
    • Accessibility
    Discuss this: Human motion and sign-language generation

Ways to work together

Research collaboration Joint projects, shared benchmarks, co-supervision.

Talks & reading groups Happy to present MAMBA, STPN and Neural Sign Actors.

Students & visitors PhD applicants and visiting researchers at QUB.

School of EEECS, Queen’s University Belfast, Belfast, UK