Closed-loop training of organoids to balance a Cartpole (Inverted Pendulum)

The recent publication by Ash Robbins and colleagues at the University of California Santa Cruz demonstrated that cortical organoids could learn to improve their performance in a cartpole (inverted pendulum) balancing task. Organoids were plated to grow continuously on MaxWell Biosystems’ MaxOne Single-Well High-Density Microelectrode Array (HD-MEA) chips, where neurons can be flexibly recorded and stimulated with high spatial-temporal resolution. The firing activity of individual neurons within the organoid encodes the positions of the pole and synpatically connected ones acts to balance it.

Adaptative learning (top) through the closed-loop electrical stimulation of targeted neurons improved the performance of organoids in balancing the cartpole. On the other hand, neither random (middle) nor null (bottom) stimulation yielded any improvement in performance. In addition, the Robbins and colleagues (2026) demonstrated that this observed short-term learning in organoids is dependent on glutamatergic transmission. Pharmacological blockade of glutamatergic transmission in organoids prevented any improvements in performance.

‍

Aside from this scientific publication, Ash – the recently named Forbes 30 under 30 in 2025 for Healthcare, co-founded the Immergo Labs and am leveraging reinforcement learning in a virtual environment for physical rehabilitation.

Mircea – a skillful and resourceful gardener outside of the lab, is iterating his little patch of land outside of his home. Check out this other impressive fruit of his labor.