MGI's subsidiary, Genoria AI, in collaboration with the Shanghai Artificial Intelligence Laboratory, announced the launch of two breakthrough innovations that close the gap between digital intelligence and physical execution in biology: ProtoPilot, a self-evolving multi-agent system driven by real-world laboratory scenarios; and BioLab Bench, the industry's first comprehensive evaluation framework that assesses AI agents from user requirements to executable device operations.

Together, these innovations establish a new paradigm—Physical AI for life sciences—where intelligent agents do not merely generate textual answers but translate experimental intent into physically executable, verifiable, and reproducible actions on automated lab platforms. The research behind it was published as a preprint on arXiv (arXiv:2606.31763) in June 2026.

ProtoPilot: A Full-Chain Agent System That Learns from Failure

ProtoPilot is a self-evolving multi-agent system that covers the entire experimental lifecycle:
Design2Protocol → Protocol2Code → Device Execution → Wet-Lab Feedback

It learns from failure. When a PCA assembly step failed, ProtoPilot diagnosed the issue (antibiotic resistance screening failure) and autonomously regenerated a corrected protocol—proving true Physical AI is here.

On ProtocolQA, one of the most representative public benchmarks for evaluating AI experimental reasoning capabilities (built by AI4S leader Future House):

GPT-5.6-sol scored 43.5 per cent

Human expert level stands at 54 per cent

ProtoPilot achieved 52.38 per cent —approaching expert-level performance!

Key features include:

Real-World Task Coverage: BioLab Bench spans from fundamental operations to complex multi-step workflows, stratified across three difficulty levels (L1–L3).

Full-Chain Assessment: Rather than merely checking whether an agent generates a plausible protocol, BioLab Bench evaluates each step—intent interpretation, protocol design, device-agnostic SOP generation, device-specific SOP translation, machine code production, and successful execution gate verification.

Cross-Device Transferability: The benchmark can be deployed on different automated laboratory platforms to test whether an AI agent can comprehend experimental tasks and generate executable actions adapted to varying hardware configurations, thus assessing cross-device generalisation capability.