Today, Eastworlds unveiled the Unitree G1, an autonomous robot that can reliably pick up a bottle, achieving this feat with a remarkably low training cost of just $200. This breakthrough was made possible by developing a model that operates without a vision-language model (VLM) backbone, which typically requires significantly more computational resources. Eastworlds, an initiative under Virtuals Protocol, is committed to open-sourcing its full dataset, allowing others to verify and expand upon their work in physical agent development. To ensure high data quality, Eastworlds employs a double-blind quality assurance process wherein trajectories are independently assessed by both teleoperators and a separate QA team.

Eastworlds: Eastworlds is an initiative dedicated to advancing physical AI agents by developing data platforms that address bottlenecks in robotics training. It operates as part of Virtuals Protocol and focuses on cost-efficient model development along with open-sourcing resources. In this announcement, Eastworlds demonstrated its approach through a Unitree G1 robot trained for reliable autonomous bottle pickup without relying on expensive VLM backbones.
Unitree G1: The Unitree G1 is a humanoid robot platform featured in Eastworlds’ demonstration of autonomous manipulation capabilities. It was trained using a cost-effective model to reliably perform tasks like picking up a bottle independently. This application highlights Eastworlds’ mission to make physical agent development more accessible via open-sourced, high-quality trajectory data.
Virtuals Protocol: Virtuals Protocol serves as the parent organization supporting specialized initiatives like Eastworlds in the physical AI space. It enables projects aimed at overcoming data limitations for agents through structured data collection and quality processes. The protocol is directly tied to the recent unveiling of Eastworlds’ low-cost training breakthrough and open dataset release.

Open Data Mission: Eastworlds prioritizes open-sourcing full datasets to allow others to verify and build upon their work in physical agent development.
Training Efficiency: The Eastworlds model achieved reliable autonomous manipulation without a VLM backbone, significantly lowering the compute requirements compared to typical approaches.
Data Quality Assurance: Eastworlds implements a double-blind QA process where teleoperators and a separate team review trajectories to ensure only high-quality data enters the dataset.