en.Wedoany.com Reported - Genoria AI, a subsidiary of MGI, and the Shanghai Artificial Intelligence Laboratory jointly announced the launch of two innovations—ProtoPilot and BioLab Bench—aimed at bridging the gap between digital intelligence and biological physical execution.

Together, these two innovations establish a new paradigm for physical AI in life sciences, where intelligent agents not only generate textual answers but also translate experimental intent into physically executable, verifiable, and reproducible operations on automated experimental platforms. The research findings were published as a preprint on arXiv (arXiv:2606.31763) in June 2026.

ProtoPilot is a self-evolving multi-agent system covering the entire chain from design to protocol, protocol to code, device execution to wet-lab feedback. The system learns from failures; for example, after a failed PCA assembly step, it diagnosed an antibiotic resistance screening issue and autonomously generated a corrected protocol. On the ProtocolQA benchmark—constructed by Future House, an institution in the AI4S field, to evaluate AI experimental reasoning capabilities—ProtoPilot scored 52.38%, approaching the human expert level of 54%, while GPT-5.6-sol scored 43.5%.

BioLab Bench is the first evaluation system to assess an agent's ability to perform tasks on real automated equipment. The system covers real-world tasks organized hierarchically across three difficulty levels, L1 to L3; provides end-to-end evaluation, examining every step from intent interpretation to execution checkpoint verification; and features cross-device migration capability, enabling deployment on different automated laboratory platforms to assess agent adaptability to various hardware configurations.

Looking ahead, biological agents will accumulate real research tasks, automated operations, expert verification, failure cases, and wet-lab feedback through physical AI experimental loops, generating massive amounts of physical experimental data, ultimately driving 7×24-hour unmanned intelligent laboratories.

MGI's exploration of AI dates back to 2019. In 2025, a team led by Dr. Yang Meng, Chief AI Officer of MGI, collaborated with Professor Nattiya Hirankarn from Chulalongkorn University to publish a paper in Nature Biomedical Engineering, introducing the PrimeGen dry-wet collaborative multi-agent system. In April 2026, MGI established Genoria AI, a subsidiary focused on AI for Science. Dr. Yang Meng noted that this path differs from pure computing power races reliant on scale, instead leveraging agent scaling and closed-loop data engineering, utilizing real tasks, device constraints, expert feedback, and wet-lab results to enable continuous AI evolution.

Genoria AI is a subsidiary of MGI, dedicated to building self-evolving laboratories to expand agent discovery. The Shanghai Artificial Intelligence Laboratory was inaugurated at the World Artificial Intelligence Conference in July 2020, positioned as a national-level new-type research institution.