Researchers at the CBoRG Lab within the Munich Institute of Robotics and Machine Intelligence (TUM MIRMI) develop robotics and AI systems tailored for biological laboratories, an approach they call “Embodied Laboratory Intelligence.” 

Standardised analysis of cancer cells 

“We aim to enable experts to engage with an intelligent laboratory, where experiments are autonomously planned, carried out, and analysed in a standardised manner while humans remain continuously involved”, explains researcher Dennis Knobbe from the CBoRG Lab. Therefore, the TUM researchers engineered and deployed robot assistants in the laboratory as human helpers. As virtual agents, they analyse data; as robotic manipulators, they perform repetitive tasks such as pipetting, cleaning and sample transport; and automated microscopes document cell imaging in detail. Here is how the whole process works in detail: 

Step 1: Establish and perturb the cell model 

Homogeneous cancer cell lines are first cultured in standardised laboratory environments to create a controlled and reproducible baseline system. Defined perturbations such as drug treatments or mechanical, electrical, and environmental stressors are then systematically introduced to initiate measurable cellular responses. So-called cooperative robotic assistants support repetitive laboratory tasks, including pipetting, media preparation, sample transport, and experimental scheduling. At the same time, human researchers supervise objectives, experimental design, and critical decision points, as well as steps in complex expert tasks. 

Step 2: Capture single-cell responses to perturbation 

After that, networked laboratory systems regularly acquire autonomously multimodal experimental data across multiple conditions and time points. Automated imaging platforms capture changes in cell growth, proliferation, cell death, morphology, migration, and invasion behaviour in single cells in a standardised and reproducible manner. 

Step 3: Identify functional variability and rare cell states 

The resulting single-cell datasets are then continuously processed by AI-driven analysis pipelines and virtual data agents to detect functional heterogeneity between cells. This enables the identification of rare subpopulations associated with resistance, persistence, invasiveness, or altered growth behaviour. Standardised data integration further allows a precise analysis across experiments and conditions. 

Step 4: Track dynamic cellular state transitions over time 

Once functional variability and rare states have been identified, the cells are continuously monitored autonomously. This is how it can be reconstructed, how cells adapt and transition between states during treatment or environmental change. Intelligent laboratory systems coordinate experimental execution, longitudinal data acquisition, and temporal analysis to identify emerging adaptive behaviours and trajectories. 

Step 5: Infer critical control systems and signalling networks 

These dynamic cellular trajectories are subsequently integrated with AI-augmented mechanistic models to infer the signalling pathways, regulatory mechanisms, and control systems that drive cellular adaptation and tumour-associated functional changes. This step transforms observational data into a mechanistic understanding of the processes. 

Step 6: Prioritise and validate intervention strategies 

Based on the identified control mechanisms and candidate vulnerabilities, therapeutic intervention strategies are prioritised and experimentally validated. Robotic assistants cooperatively execute validation workflows, while AI systems evaluate outcomes across experimental conditions and rank the most promising targets for expert review. 

Step 7: Iteratively refine hypotheses through autonomous experimentation 

These findings then feed back into an AI-guided experimental cycle in which the hypotheses will be regularly refined, and the experimental design further optimised. Human researchers remain actively integrated throughout the process, ensuring a synergistic collaboration between scientists, virtual AI agents, and embodied robotic laboratory assistants. 

Text: Andreas Schmitz