View Proposal

---

Proposer

Alessandro Suglia

Title

NaturalOXE: realistic natural language instructions for robot learning

Goal

Create true natural language instructions for robotics tasks

Description

Open X-Embodiment is a recent benchmark for robotics tasks that was created to push the boundaries of robot learning. This benchmark contains trajectories for different embodiments and language instructions. However, natural language instructions are very simple and far from being realistic. As we've seen with ChatGPT and LLMs more broadly, it is essential that the next generation of robots is able to understand human commands. In this project, you will extend the Open X-Embodiment benchmark to create realistic natural language instructions associated with each task in the benchmark. This will represent the starting point of a bigger project aiming at studying models' robustness to complex natural language instructions for robotics tasks.

Resources

https://robotics-transformer-x.github.io/

Background

Open X-Embodiment: https://robotics-transformer-x.github.io/ https://arxiv.org/abs/2407.03967

Url

Difficulty Level

Moderate

Ethical Approval

None

Number Of Students

1

Supervisor

Alessandro Suglia

Keywords

natural language processing,embodied ai,robotics

Degrees

Bachelor of Science in Computer Science
Master of Science in Artificial Intelligence
Master of Science in Human Robot Interaction
Master of Science in Robotics
Master of Science in Robotics with Industrial Application