Research question

How should sensing, reasoning and execution be divided between network-side models and the intelligence on each robot when both communication and computation are constrained?

Approach under investigation

Large models running on the network side can reason across several robots, but they reach the physical world only through wireless links and the robots’ own controllers. We investigate how to split the work so that each part runs where it fits best.

  • Division of labour between network-side coordination and on-robot execution, with time-critical control kept local to each robot.
  • Shared evidence — which observations and task states robots exchange, and how the network side maintains a common picture of the task.
  • Communication constraints — how bandwidth, delay and reliability shape what can be coordinated remotely.
  • Multi-robot decisions across heterogeneous humanoid and quadruped platforms.

Diagrams and media

Concept: a network-side coordination node exchanges shared evidence with several robots, each running its own local intelligence.

  1. Network-side coordination Shared evidence and task allocation
    • Robot A Local perception, decisions, control
    • Robot B Local perception, decisions, control
    • Robot C Local perception, decisions, control

↺ Local observations and evidence return to the coordinator

Concept illustration

Network-side coordination with on-robot intelligence

Concept diagram of the approach under investigation, not an experimental result.

Related projects

All research

Ongoing research Whole-Body Control & InteractionDistributed Embodied Intelligence

Natural Interaction with Mobile and Humanoid Robots

We investigate interfaces that connect human instructions with robot behaviours, supporting interactive demonstrations across mobile and humanoid platforms.

Project investigator: Haonan Zhou