Agentic AI for Humanoid Social Robots (GINNY Project)

Primary supervisor

Hamid Rezatofighi

Co-supervisors

Want to work with a real humanoid robot?

GINNY is a humanoid robotics project at Monash FIT exploring how robots can become more capable, adaptive, and socially intelligent partners.

You’ll work at the intersection of robot perception, LLMs, planning, and neuro-symbolic reasoning, building systems that allow a humanoid to understand its environment, reason about people and relationships, and make explainable decisions.

The key research question is simple but ambitious:

Can a humanoid learn from its own mistakes, recover when things go wrong, and become better at helping people over time?

If you're interested in humanoid robotics, embodied AI, agentic systems, or LLM-powered robots, this project gives you the chance to develop and test ideas on a real robotic platform.

Aim/outline

  1. Design an interactive agent architecture for humanoid robots (Pepper, Unitree G1) grounded in perception.
  2. Implement task execution so the humanoid can carry out real-world actions.
  3. Build a recursive learning loop so the agent improves from failed interactions.
  4. Evaluate on a live humanoid platform and write up the thesis.

URLs/references

https://arxiv.org/abs/2605.00943

ARIS: Agentic and Relationship Intelligence System for Social Robots

For any inquiries regarding this project, please contact Stavya Datta via stavya.datta@monash.edu 

Required knowledge

  • Python programming (C/C++ a bonus)
  • Basic computer vision and ML/DL concepts
  • Linux DevOps
  • Software engineering basics
  • ROS a bonus