Scalable Low-Cost Laboratory Automation: A Digital Twin-Integrated Robotic Platform for Autonomous Liquid Handling (RAINBOT)

Technical Summary: Scalable Low-Cost Laboratory Automation via RAINBOTTM

Problem Statement
While self-driving laboratories (SDLs) offer a powerful route to accelerate discovery in chemistry and materials science by closing the loop between hypothesis, experiment, and analysis, their adoption is currently limited by three primary barriers: the high cost of commercial liquid-handling systems, their proprietary nature, and the lack of remote supervisability. Existing low-cost, open-hardware solutions often lack a live, synchronized representation of the experiment, preventing remote experts from monitoring or intervening in real time. Conversely, high-fidelity digital twins and closed-loop autonomous systems typically exist as computationally intensive simulations or expensive, closed-unit instruments that do not integrate with inexpensive, modifiable hardware.

Methodology
The authors present RAINBOTTM (Real-time Autonomous Intelligent Network for Bidirectional Optimization and Traceability), a low-cost, openly reproducible liquid-handling robot. The system integrates three distinct capabilities rarely combined in a single platform:

1. Hardware Architecture: The core motion platform is a modified consumer-grade Cartesian 3D printer (Elegoo Neptune 4 Max). The extruder is replaced by a carriage holding a precision single-channel pipette (Eppendorf Research plus, 100–1000 µL). Aspiration and dispensing are actuated mechanically by two compact linear actuators (30 mm and 50 mm strokes) that depress the pipette's native plunger and tip-eject buttons, respectively. A GY-33 color sensor (TCS34725 RGB module) is mounted above a magnetically stirred mixing vessel to provide real-time colorimetric feedback. The total hardware cost is approximately US$1,263.
2. Software and Digital Twin: A custom Python library orchestrates motion by streaming G-code to the printer's Klipper firmware and controlling actuators via a USB relay board. A browser-based digital twin, implemented as a WebGL build of a Unity project, synchronizes bidirectionally with the physical platform. It mirrors the X–Y–Z gantry kinematics and pipetting states in real time, allowing remote monitoring, intervention, and emergency stopping from any web browser. An onboard webcam provides an independent visual log of the workspace.
3. Closed-Loop Control: The platform is coupled with the CEIDTM (Cooperative Explorer for Inverse Design) framework. This system recasts experimentation from iterative manual guessing to a goal-directed inverse-design search. CEIDTM takes the current colorimetric state and measurement history as input to propose the next pipetting action (volumes of red, yellow, blue, and water) to minimize the distance between the measured spectrum and a user-specified target. A human operator remains in the loop to monitor or override the system at any stage.

Proof-of-Concept Experiment
The system was validated using a benign, fully observable color-mixing task. The robot performed sequential exchanges of red, yellow, and blue aqueous dye solutions. The integrated color sensor quantified the resulting mixtures, and the system executed a closed-loop optimization to match a target spectrum defined by an eight-channel AS7341 reference.

Key Results

• Cost and Accessibility: The complete platform costs under US$1,300, roughly an order of magnitude cheaper than entry-level commercial liquid handlers (e.g., Opentrons OT-2 at ~$5,000) and significantly below high-end commercial systems (>$50,000).
• Dispensing Accuracy: Gravimetric validation at nominal volumes of 200, 500, and 1,000 µL demonstrated high precision. The coefficients of variation (CV) were 0.303%, 0.203%, and 0.182%, respectively, confirming sub-0.5% repeatability.
• Sensor Fidelity: During color-mixing trials, the measured red, yellow, and blue (RYB) channel responses agreed with expected values based on dispensed volumes to within a mean absolute error of two percentage points.
• Optimization Performance: Using the CEIDTM framework, the system successfully identified an optimal formulation (5 red, 1 yellow, 5 blue, 2 water units) at trial 16 out of a 24-experiment budget. The final corrected Frechet distance between the measured and target spectra was 0.014524.
• Digital Twin Synchronization: The browser-based twin mirrored physical motions with sub-second latency, enabling real-time remote supervision and immediate human intervention via a software emergency stop.

Significance and Claims
The paper claims that RAINBOTTM establishes an accessible physical–virtual framework for self-driving laboratory automation. Its primary significance lies not merely in its low cost, but in the unique integration of open, modifiable hardware with a live, remotely supervisable digital twin and goal-directed inverse-design autonomy. By keeping a human in the loop and providing a transparent, bidirectional interface, the platform addresses the "black box" nature of many autonomous systems. The authors assert that this approach lowers both the financial and supervisory barriers to laboratory automation, allowing resource-constrained settings to participate in autonomous discovery workflows previously restricted to well-funded institutions. The work demonstrates that research-grade automation can be built from modular, low-cost components without sacrificing the ability to monitor, steer, and trust the experimental process.