Data Generation & Benchmark for Instruction-Following VLN models

Duration: TBD
Affiliation: VinMotion, VinGroup · Hanoi, Vietnam

Note: Full write-up coming soon.

Overview

Developed an automated data generation pipeline for creating large-scale instruction-following datasets targeting robotic manipulation. The benchmark evaluates embodied agents on a diverse set of household and industrial tasks.

Planned Content

• Pipeline architecture for scalable instruction-following data generation
• Task diversity and difficulty distribution analysis
• Benchmark evaluation metrics and baseline results
• Integration with VLA training workflows

Technologies

Python VLM VLA LLM ROS 2 PyTorch Data Generation