Outline
This project builds an end-to-end pipeline for real-robot manipulation: **teleoperate** the robot to collect demonstrations, create a clean dataset, and train imitation policies via **fine-tuning of an open-source Vision-Language-Action (VLA) model**. Our goal for this project is to build the setup for participating in the **WBCD manipulation competition** (What Bimanuals Can Do) at IROS 2026.
Milestones
- M1 (Weeks 1–2): Hardware/ robotics setup.
- M2 (Weeks 2–6): Implement teleoperation + data collection pipeline.
- M3 (Weeks 7–10): Data collection + setup of VLA fine-tuning pipeline.
- M4 (Weeks 11-14): Fine-tune and evaluate learned policy.
Requirements
Strong Python/ML coding skills (PyTorch), ROS, experience and motivation to work with real robots.
To apply, email (1) a short paragraph on your background/fit and (2) transcripts to [email protected], [email protected], and [email protected].
Reference
WBCD Competition Website: https://wbcdcompetition.github.io/