Building the Next Standard for Robotic Intelligence

Primary supervisor

Fucai Ke

Co-supervisors

This project focuses on developing data-driven benchmarks for robotic intelligence. The research will identify important limitations in current robotic systems, translate these limitations into measurable research questions, and design datasets and evaluation protocols that systematically test them.

Students will investigate how to discover meaningful benchmark problems from real robotic scenarios, design challenging and reproducible tasks, collect and annotate robotic data, define evaluation metrics, and benchmark state-of-the-art robotic and multimodal models. The project can cover a broad range of capabilities, including perception, spatial and temporal reasoning, manipulation, planning, and long-horizon interaction.

Aim/outline

  • Identify fundamental limitations and open problems in current robotic systems.
  • Formulate these problems into well-defined and measurable benchmark tasks.
  • Design and collect data-driven robotic datasets for systematic evaluation.
  • Develop robust evaluation metrics and protocols for robotic intelligence.
  • Benchmark state-of-the-art robotic, vision-language, and embodied AI models.
  • Establish new evaluation standards for future robotic systems.

URLs/references

VIEW2SPACE: Studying Multi-View Visual Reasoning from Sparse Observations, ECCV2026 — the main reference paper for the project.

https://arxiv.org/abs/2603.16506

For any inquiries regarding this project, please contact Fucai Ke via fucai.ke@monash.edu

Required knowledge

  • Strong willingness and ability to continuously learn new concepts and technologies.
  • Ability to commit consistent time to the project every week.
  • Good programming skills in Python.
  • Basic knowledge of machine learning, computer vision, or robotics.
  • Experience with PyTorch, vision-language models, embodied AI, or robotic systems is desirable.
  • Ability to work independently and conduct research systematically.