RoboCulture: A Robotics Platform for Automated Biological Experimentation

TL;DR

RoboCulture presents a flexible, end-to-end robotic platform for autonomous biological experimentation, integrating a 7-axis manipulator, a vision-driven liquid handling system, force-guided tip exchange, and optical-density-based growth monitoring. The system is orchestrated by modular behavior trees, enabling reactive decision-making and long-duration operation without human intervention. A fully autonomous 15-hour Saccharomyces cerevisiae culture demonstrates end-to-end capabilities, including random-well pipetting, tip exchange, growth monitoring, and well-splitting guided by real-time perception; the authors provide open-source code and CAD models to promote adaptation and extension. Through emphasizing autonomy and adaptability over throughput, RoboCulture highlights a practical path toward generalizable, operator-free biology laboratories, with clear directions for robustness and scalability enhancements.

Abstract

Automating biological experimentation remains challenging due to the need for millimeter-scale precision, long and multi-step experiments, and the dynamic nature of living systems. Current liquid handlers only partially automate workflows, requiring human intervention for plate loading, tip replacement, and calibration. Industrial solutions offer more automation but are costly and lack the flexibility needed in research settings. Meanwhile, research in autonomous robotics has yet to bridge the gap for long-duration, failure-sensitive biological experiments. We introduce RoboCulture, a cost-effective and flexible platform that uses a general-purpose robotic manipulator to automate key biological tasks. RoboCulture performs liquid handling, interacts with lab equipment, and leverages computer vision for real-time decisions using optical density-based growth monitoring. We demonstrate a fully autonomous 15-hour yeast culture experiment where RoboCulture uses vision and force feedback and a modular behavior tree framework to robustly execute, monitor, and manage experiments. Video demonstrations of RoboCulture can be found at https://ac-rad.github.io/roboculture.

Figures (27)

• Figure 1: RoboCulture integrates a vision‐based liquid handling system capable of reliable pipetting into 96-well plates, a force-guided pipette tip exchange system, and cellular growth monitoring toward generalizable biology laboratory automation. Biology protocols are represented as behavior trees, a reactive and modular framework for experiment state handling. Code, CAD models and video demonstrations of RoboCulture can be found at https://ac-rad.github.io/roboculture/.
• Figure 2: RoboCulture bridges human-level biological protocols and low-level robotic tasks using behavior trees. High-level experimental instructions such as growth monitoring and sub-culturing yeast are translated into modular, hierarchical behavior trees composed of individual behaviors. These trees coordinate key robotic subsystems, including optical density perception, vision-based robot control, and pipetting, enabling autonomous execution and decision making during complex cell culture workflows. A more detailed description of the behavior trees and individual behaviors is provided in Section \ref{['sec:behavior-trees']}.
• Figure 3: The yeast growth experiment setup. 1) Our Digital Pipette v2, 2) a downwards-facing Intel RealSense D435i camera, 3) a Franka Emika robot with a Robotiq 2F-85 gripper, 4) an OHAUS SHHD1619DG Heavy Duty Orbital Shaker Platform, 5) a 3D printed pipette tip remover, 6) a biological waste bin, 7) a Falcon Tube rack holding YPD media, 8) a 3D printed pipette tip rack, and 9) a 96 well plate prepared with yeast.
• Figure 4: Gravimetric comparison of the Digital Pipette v2 with human pipettors, for volumes of 0.2 mL, 1 mL and 5 mL. A Levene’s test was conducted to compare the variances between the two groups (Digital Pipette v2 vs. Human).
• Figure 5: Demonstration of the spiral search process for pipette tip attachment. Initially, the pipette body and the new tip are misaligned. A spiral trajectory of waypoints in the XY plane is generated around the start position of the pipette body, and the robot moves the pipette along this trajectory while pressing the pipette onto the surface of the new tip. The force at the end-effector is monitored during the trajectory, and the pipette is finally lowered into the pipette tip when the force drops considerably.