Ultra-Low-Cost Self-Driving Laboratories: Integrating Modular Liquid Handling and AI-Based State Verification for Thin-Film Processing

Sayan Doloi¹, Amrita Joshi¹, Wong Rui Yue¹, Anushree Rawat¹, Leonard Ng Wei Tat¹

¹School of Materials Science and Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798

Self-driving laboratories can accelerate materials discovery by integrating automated experimentation, high-throughput materials processing, autonomous characterization, and data-driven decision-making [1-2]. However, their broader adoption remains constrained by the high capital cost of commercial automation platforms, proprietary system architectures, limited reconfigurability, and substantial integration complexity [3]. These barriers are particularly significant for academic and resource-constrained laboratories seeking to automate diverse experimental workflows without investing in multiple application-specific instruments. A further challenge arises from legacy and low-cost laboratory instruments whose operational states are communicated through visually readable settings, indicators, and displays that are not digitally accessible, thereby limiting their integration into closed-loop autonomous workflows.

Here, we present an ultra-low-cost (< USD 500), modular, and open-source automation framework for solution-processed thin-film research. The platform integrates automated liquid handling, thin-film deposition, robotic sample transfer, thermal processing, imaging, and UV–visible optical characterization within a reconfigurable experimental workflow. The system repurposes the Cartesian motion architecture, stepper motors, control electronics, and programmable motion capabilities of a low-cost commercial fused deposition modelling 3D printer, thereby reducing the requirement for custom robotic hardware. Using the same underlying architecture, modular workstations were developed for automated liquid handling across multiple labware formats, including vials, centrifuge tubes, and 96-well plates, together with multi-position spin-coating and drop-casting workflows for thin-film fabrication.

A key feature of the platform is a vision-language-model-based closed-loop sensing strategy that converts visually observable but digitally inaccessible instrument states into machine-readable feedback. Using an off-the-shelf camera and a pretrained vision-language model, the system autonomously reads the mechanical display of a micropipette, compares the observed state with the commanded volume, and performs corrective actuation when required. The approach generalizes across multiple pipette brands and volume ranges without task-specific retraining, demonstrating the potential of vision-language models as flexible perception layers for low-cost laboratory automation. The same perception-driven framework is further extended to image-based inspection of solution-processed thin films, where visual assessment provides an automated quality-control step for identifying macroscopic coating non-uniformity and processing defects before subsequent characterization.

These results demonstrate how inexpensive, reconfigurable hardware can be combined with AI-based visual state verification and modular robotic processing to support increasingly autonomous materials workflows. By integrating low-cost automation, vision-language-model-based perception, robotic sample transfer, and optical screening, the platform provides an accessible foundation for closed-loop optimization of solution-processed functional materials and illustrates a practical pathway toward democratizing self-driving laboratories.

Keywords: self-driving laboratories; laboratory automation; vision-language models; closed-loop control; thin-film processing; high-throughput experimentation.