Research question

How can spoken or typed instructions be connected to robot behaviours in a way that people can predict and rely on?

Approach under investigation

People should be able to ask a robot for something in their own words. We investigate the interface layer between those words and what a robot can actually do.

  • Intent from speech and text — interpreting what a person is asking the robot to do.
  • Behaviour interfaces that link interpreted intent to the behaviours a platform supports.
  • Interactive demonstrations across mobile and humanoid platforms.

Diagrams and media

Concept: spoken or typed instructions are interpreted as intent and mapped through a behaviour interface to mobile and humanoid platforms.

    • Speech
    • Text
  1. Intent interpretation What is being asked
  2. Behaviour interface Map intent to supported behaviours
    • Mobile platforms
    • Humanoid platforms

↺ Robot responses inform the next instruction

Concept illustration

From instructions to robot behaviours

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

Physical robot Unitree Go2

Voice and semantic control of two Go2 quadrupeds

Spoken instructions direct two Go2 quadrupeds, named Jarvis and Micros, through a voice and semantic control interface. Each robot answers “Copy that” and carries out the command. With sound and captions.

What this shows: Jarvis is told to stand down, stand up and do a backflip; Micros is told to do a backflip and then sit down. Each command is acknowledged before the robot acts.

Investigator: Haonan Zhou

Recorded 31 May 2026

Concept illustration: a spoken or typed instruction is interpreted as intent, mapped to a robot behaviour and carried out by a Unitree Go2 quadruped.

  1. Instruction Spoken or typed
  2. Intent Interpreted request
  3. Behaviour Selected and carried out

↺ The person sees the response and can follow up

Concept illustration Unitree Go2

Go2 human–robot interaction

Research direction: interfaces that turn spoken or typed instructions into behaviours on a quadruped platform.

Investigator: Haonan Zhou

Related projects

All research

Ongoing research Distributed Embodied Intelligence

Collaborative Intelligence Across Robots and Networks

We study how network-side models and on-robot intelligence can coordinate sensing, planning and execution across multiple robots, while accounting for communication and computation constraints.

Project investigator: Haonan Zhou

Ongoing research Whole-Body Control & Interaction

Balance-Aware Whole-Body Manipulation

We explore how whole-body dynamics and geometric constraints can inform upper-body motion, with the aim of improving stability during humanoid manipulation.

Project investigator: Haonan Zhou