Hong Kong Researchers Deploy AI-Driven Robotic Boxes to Accelerate Perovskite Solar Cell Discovery

Hong Kong researchers use an agentic robotics system and a Recipe Language Model to automate over 50,000 perovskite solar cell experiments.

By: AXL Media

Published: Apr 14, 2026, 7:35 AM EDT

Source: Information for this report was sourced from EurekAlert

Automating the Future of Photovoltaics

Perovskite solar cells represent a critical frontier in next-generation energy, yet their development has long been hindered by labor-intensive trial-and-error synthesis. Researchers from the Hong Kong Polytechnic University and their collaborators have addressed this bottleneck by introducing an agentic robotics system designed specifically for perovskite research. According to a report published in Engineering in 2026, the framework replaces fragmented manual operations with a unified, automated environment. By integrating a language agent with robotic execution, the team has moved beyond simple automation toward a system capable of semantic reasoning and feedback-driven discovery.

The Seven-Layer AI Architecture

At the heart of this technological leap is a sophisticated seven-layer artificial intelligence architecture. This framework encompasses learning, generating, fine-tuning, and reasoning, allowing the system to process both numerical data and semantic information from scientific literature. The researchers developed a domain-specific Recipe Language Model (RLM) that encodes complex chemical formulas into machine-readable instructions. This RLM continuously improves by learning from its own experimental outputs and existing corpora, creating a closed-loop workflow where the AI recommends, validates, and refines its own solar cell recipes.

A Modular Robotic Environment

The physical infrastructure of the system consists of 11 interconnected robotic boxes that manage every stage of the solar cell lifecycle, from initial synthesis to final characterization. This hardware setup includes 101 functional modules and over 1,500 individual components, all controlled by more than 4,300 unique parameters. To manage this complexity, the team utilized a digital twin, a real-time software interface that synchronizes the physical state of the robots with the AI's model-generated instructions. This setup ensures that the extremely sensitive crystallization process of perovskites remains stable and controllable within an enclosed environment.