A Closed-Loop Robot Scientist for Autonomous Biological Discovery
- 1Tufs University
Abstract
Biological systems respond to a wide range of physical inputs and an overarching goal of biological research is understanding how these signaling events coordinate cellular, tissue, and organism-level outcomes. Yet the vast combinatorial space of physical and chemical interventions that influence these processes remains largely unexplored. Thus, a search process is required that efficiently learns how inputs affect a biological target via a series of automatically generated interventions, and a robot scientist that can conduct them. To this end we here introduce the Multimodal Organismal Modulation Robot (MOMbot), a robot scientist that integrates four physical intervention modalities - chemical delivery, electrical field application, mechanical vibration, and thermal modulation -within a single hardware platform. MOMbots multimodal hardware is paired with high-resolution imaging and an online active learning algorithm that autonomously generates, executes, and refines interventions based on accumulated data. We validate the system across biological materials spanning three orders of magnitude in scale, including Xenopus laevis embryos, motile mucociliary organoids, and disaggregated ectodermal stem cells, to show the utility of this approach across diverse biological disciplines. Using these biological models, we demonstrate precise thermal control of embryonic developmental rate, automated chemical perturbation with behavioral readouts, programmable electrical field stimulation, and vibrational assembly of functional mucociliary organoids from loose cells. We also describe the active learning algorithm that autonomously concentrates interventions near regions of maximal outcome uncertainty, yielding sample-efficient discovery of an electrical duration-response relationship for motile organoids. These results establish MOMbot as a scalable robot/AI scientist team for biological discovery, enabling autonomous, multimodal physical experimentation and AI-driven exploration of complex biological systems.
Related
- ARSA: an autonomous research scientist for target nomination in Alzheimer's disease and related dementias
- AutoScreen: AI Co-Scientist System for Target Discovery in Functional Genomics
- An AI-powered cloud biofoundry for autonomous biological research
- Research note: approach rather than avoidance: behavioral responses of broiler chicks repeatedly exposed to an autonomous mobile robot
- Edge-deployable greenhouse tomato cluster harvesting robot integrating YOLOv8n-BiFPN-WIoU and ROS-based autonomous control
BibTeX
@misc{bielawski2026closedloop,
title = {{A Closed-Loop Robot Scientist for Autonomous Biological Discovery}},
author = {K. Bielawski and K. Srinivasan and N. Gaylinn and S. Beaulieu and R. Brucker and M. Levin and J. Bongard and D. Blackiston},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.11.751076},
url = {https://www.biorxiv.org/content/10.64898/2026.09.11.751076v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - A Closed-Loop Robot Scientist for Autonomous Biological Discovery
AU - K. Bielawski
AU - K. Srinivasan
AU - N. Gaylinn
AU - S. Beaulieu
AU - R. Brucker
AU - M. Levin
AU - J. Bongard
AU - D. Blackiston
PY - 2026
DA - 2026/09/11
PB - bioRxiv
DO - 10.64898/2026.09.11.751076
UR - https://www.biorxiv.org/content/10.64898/2026.09.11.751076v1
AB - Biological systems respond to a wide range of physical inputs and an overarching goal of biological research is understanding how these signaling events coordinate cellular, tissue, and organism-level outcomes. Yet the vast combinatorial space of physical and chemical interventions that influence these processes remains largely unexplored. Thus, a search process is required that efficiently learns how inputs affect a biological target via a series of automatically generated interventions, and a robot scientist that can conduct them. To this end we here introduce the Multimodal Organismal Modulation Robot (MOMbot), a robot scientist that integrates four physical intervention modalities - chemical delivery, electrical field application, mechanical vibration, and thermal modulation -within a single hardware platform. MOMbots multimodal hardware is paired with high-resolution imaging and an online active learning algorithm that autonomously generates, executes, and refines interventions based on accumulated data. We validate the system across biological materials spanning three orders of magnitude in scale, including Xenopus laevis embryos, motile mucociliary organoids, and disaggregated ectodermal stem cells, to show the utility of this approach across diverse biological disciplines. Using these biological models, we demonstrate precise thermal control of embryonic developmental rate, automated chemical perturbation with behavioral readouts, programmable electrical field stimulation, and vibrational assembly of functional mucociliary organoids from loose cells. We also describe the active learning algorithm that autonomously concentrates interventions near regions of maximal outcome uncertainty, yielding sample-efficient discovery of an electrical duration-response relationship for motile organoids. These results establish MOMbot as a scalable robot/AI scientist team for biological discovery, enabling autonomous, multimodal physical experimentation and AI-driven exploration of complex biological systems.
ER -