Neural Noise- Microelectrode arrays (MEAs) cultured with in vitro neural networks are used as a testbed for encoding and decoding relevant to motion control.1
- Progress and advances in this paradigm translate to invasive BCI R&D and neuroprosthetics; the work was reported in Nature and is tier-1.1 1
Weekly enrichment (2026-07-20)
- The underlying source is a 2025 review in Microsystems & Nanoengineering (Nature portfolio; Hua et al., vol. 11, no. 1), not an original experiment; it surveys “bio-integrated systems” that couple MEA-cultured in vitro neural networks to a bidirectional communication layer and a physical or simulated actuator for closed-loop motion control.2
- Decoding methods reviewed span classical firing-rate (FR) mapping and central-firing-rate (CFR, population-coding-inspired) algorithms through AI approaches — reservoir computing, deep neural networks, and spiking-neural-network-inspired models; the review concludes low-complexity algorithms suit fast, simple decisions while deeper models better handle complex temporal and dynamically changing tasks.2
- On encoding, the review contrasts rate coding versus temporal coding and notes most motion-control systems use rate coding because temporal coding’s information gain is outweighed by its resource cost.2
- Stimulation is typically delivered as charge-balanced biphasic rectangular pulses with amplitudes generally 100–500 mV (occasionally exceeding 1 V under stress-test conditions); reward/punishment schemes (a fixed reward pattern that lowers network entropy versus randomized punishment) exploit synaptic plasticity to drive learning.2
- Cultures are usually primary neurons from rat embryos/neonates or human induced pluripotent stem cells (hiPSCs); the review cites evidence that human cultures sustain correct gameplay longer, while hiPSC variability risks tumor-cell proliferation that alters network connectivity.2
- Modular culture techniques (e.g., PDMS molds constraining connectivity) can enhance directional information flow up to 100-fold, improving performance on intelligent tasks.2
- Landmark systems surveyed include Kagan et al.’s DishBrain playing “Pong,” Cai et al.’s 3D brain-organoid reservoir computing for speech recognition and nonlinear-equation prediction (Nat. Electron. 6, 1032–1039, 2023), Bakkum’s MEART robotic-arm drawing, and Yada et al.’s FORCE reservoir computing in living cultures.23
- Earlier robot-control work established the closed-loop culture-to-robot paradigm this review builds on: Warwick et al. drove a Miabot for obstacle avoidance using rat-neuron cultures on a 64-site (8×8) MEA, and DeMarse et al. controlled a simulated aircraft’s pitch and roll from stimulation-evoked responses.3
- Hardware is trending toward 3D and high-density MEAs (e.g., a 4096-channel event-based array) with FPGA/GPU acceleration for real-time decoding; scaling to tens of thousands of electrodes and 3D organoids is flagged as a real-time decoding bottleneck.24
- Clinical/BCI implication: these advances feed invasive-BCI and neuroprosthetics R&D and point toward ultra-low-power “bio-computers,” while the review also raises ethics concerns over potential organoid consciousness and donor rights.2