Physical intelligence (PI) fundamentally differs from computational intelligence. While computational systems process information through electronic circuits in a central processor, physical intelligence embeds sensing, actuation, memory, and even logic into the material and structural design of a body itself.1
As Metin Sitti of the Max Planck Institute articulated in Extreme Mechanics Letters, PI can be defined as physically encoding sensing, actuation, control, memory, and decision-making into the body of an agent. Biological evolution has been doing this for hundreds of millions of years, and at scales where centralized neural control alone would be too slow or too costly.1
The Octopus as a Distributed Computing System
The octopus has roughly 500 million neurons in total, but only about 40% of them sit inside its central brain. The remaining 60% are distributed throughout its eight arms, with each arm running its own axial nerve cord and between 200 and 300 suckers staggered along its ventral surface.2
This decentralized arrangement means that an octopus arm can reach, grasp, and respond to local stimuli completely independent of the brain's direct instruction. When an octopus explores a crevice, one arm can simultaneously work on prying open a shell while the central brain focuses on something else entirely. This multi-scale division of labor offers a practical biological blueprint for distributed AI systems.3
Hierarchical Suction Intelligence
A recent study published in Science Robotics demonstrated that researchers have successfully “translated” an octopus’s neuromuscular hierarchy into a functioning robotic architecture. By coupling suction flow with local fluidic circuitry, soft robots could achieve low-level embodied intelligence, including gentle grasping of delicate objects, adaptive curling, and encapsulation of objects of unknown geometry.4
Critically, the system decoded pressure responses from individual suction cups to enable high levels of multimodal perception, including contact detection, surface-roughness classification, and pulling-force prediction. As with the actual octopus, computations occurred mostly locally, with only minimal information reaching the centralized decision-control layer. This architecture reduces computational overhead by orders of magnitude while maintaining dexterous performance.4
Mathematical Model of the Octopus Arm
One of the trickier engineering problems has been modeling how an octopus arm actually moves. To that end, researchers at the University of Illinois Urbana-Champaign addressed this by expressing arm musculature using a stored energy function borrowed from continuum mechanics theory. Muscle activations shift the arm's equilibrium position by modifying the stored energy, guiding motion without requiring exhaustive real-time computations.5
The model produced remarkably lifelike simulations of three-dimensional arm reaching and grasping in the Elastica software environment. Unlike machine learning approaches, this method provides mathematical performance guarantees. The team argues that such guarantees are often absent in alternative data-driven methods, making this framework both more interpretable and more reliable for robotic control design.5
Learnings About Flight and Navigation
Insects solved the problem of powered flight approximately 350 million years ago, and they did so with brains containing fewer neurons than some computers have transistors. Their secret lies in embedding aerodynamic intelligence directly into wing morphology and flight mechanics.
A study published in Frontiers in Robotics and AI showed that natural fliers use passive wing-stroke dihedral mechanics to automatically steer into wind gusts, achieving gust resistance without any active sensing or neural command.6
This passive stability mechanism inspired micro-air-vehicle (MAV) designs that maintain flight stability in turbulent conditions without relying on feedback control loops. Insects accomplish this through precisely tuned wing compliance and geometry, not through computation. For roboticists building insect-scale fliers, replicating this physical passivity substantially reduces control complexity and enables further miniaturization.1
Optic Flow and Insect Navigation
Insects also navigate using a visual strategy called optic flow regulation. When flying forward, the image of the ground sweeps backward across their visual field at a rate that encodes both speed and altitude. Research dating back to foundational bio-robotics work showed that a micro-helicopter equipped with a bio-inspired optic flow sensor could land safely, avoid the ground, and regulate altitude, all without any onboard altimeter, GPS, or speed sensor.7
This control scheme explains behaviors that puzzled entomologists for decades, such as why honeybees descend into headwinds or drown over mirror-smooth water where optic flow signals disappear. The elegance lies in how the insect converts a complex estimation problem into a single, measurable visual variable. Drone engineers have since adopted this principle to build lightweight autonomous vehicles that navigate without traditional sensor suites.7
Ant Colonies and Emergent Problem-Solving
Individual ants have relatively small nervous systems, yet colonies collectively excavate intricate tunnel networks, organize foraging routes, and respond dynamically to damage. Harvard researchers studying cooperative excavation identified just two governing parameters behind this behavior: the strength of cooperation and the rate of individual excavation. What makes this finding significant is that no individual ant needed awareness of the larger structure being built.8
Their robotic analogs, called RAnts, operated on three simple local rules: follow a chemical gradient, avoid other robots where the signal is high, and pick up obstacles where concentration is high while dropping them where it is low. These minimal rules produced complex, coordinated construction behavior without any global plan or centralized direction.8
The researchers described this as the interplay between simple individual rules and the collective's embodied physics, a hallmark of physical intelligence operating at the group level. For engineers, this means that scaling a robotic system does not always require scaling the intelligence of each unit, just the quality of local rules and environmental feedback.8
From Biology to Bionic Materials
The translation from biological observation to engineered systems depends heavily on materials science. Research in bioinspired soft robotics has grown sharply, with related publications rising from 760 in 2021 to 1,170 in 2024, a nearly 54% increase over five years.2
Actuator design for octopus-inspired robots now spans pneumatic systems modeled on arm musculature, hydraulic systems inspired by fluid-driven tentacle motion, and shape memory alloy actuators that simulate longitudinal muscle fibers running parallel to the arm axis. Each approach captures a different aspect of the octopus's physical intelligence, from the compliance of soft tissues to the precision of local neural control.2
Researchers are also exploring fiber-reinforced elastomers that mimic the anisotropic stiffness of octopus skin. With such materials, robots can deform freely and resist loads directionally, a mechanical property that conventional rigid materials cannot offer.2
Medical and Industrial Applications
The use of bio-inspired physical intelligence in medical engineering has already led to functional prototypes. One design, based on an octopus suction-cup structure, created a drug-release soft robot that can attach to stomach lesion sites using magnetic gradient force and deliver medication slowly over time. The attachment mechanism requires no active clamping, relying instead on instantaneous suction generated by the magnetic gradient itself.2
In broader industrial terms, suction intelligence systems could be integrated into fluidic-driven soft robots for agricultural harvesting, object handling in manufacturing, and interventional healthcare, all while reducing computational requirements compared to sensor-heavy conventional robotic systems.
Saving this for later? Download a PDF here.
The principles of hierarchical, locally executing intelligence are applicable wherever robustness, adaptability, and low power consumption are design priorities. As fabrication techniques improve, the gap between biological prototypes and deployable machines will narrow, and the animals that inspired these systems will continue to offer design guidance that no purely engineered approach has matched.4
References and Further Reading
- Sitti, M. (2021). Physical intelligence as a new paradigm. Extreme Mechanics Letters, 46, 101340. DOI:10.1016/j.eml.2021.101340. https://www.sciencedirect.com/science/article/pii/S2352431621001012
- Duan, J. et al. (2025). Learning from Octopuses: Cutting-Edge Developments and Future Directions. Biomimetics, 10(4). DOI:10.3390/biomimetics10040224. https://www.mdpi.com/2313-7673/10/4/224
- Shaw, A. D. et al. (2026). The Embodied Octopus: Distributed Intelligence and Active Inference in a Flexible Body. CPNS Lab, University of Exeter, Preprint. DOI:10.31234/osf.io/p7f3k_v1. https://osf.io/preprints/psyarxiv/p7f3k_v1
- Yue, T. et al. (2025). Embodying soft robots with octopus-inspired hierarchical suction intelligence. Science Robotics. DOI:10.1126/scirobotics.adr4264. https://www.science.org/doi/10.1126/scirobotics.adr4264
- Michael O'Boyle. (2023). Reaching like an octopus: a biology-inspired model opens the door to soft robot control. University of Illinois Urbana-Champaign. https://csl.illinois.edu/news-and-media/reaching-like-an-octopus-a-biology-inspired-model-opens-the-door-to-soft-robot-control
- Olejnik, D. A. et al. (2022). Flying Into the Wind: Insects and Bio-Inspired Micro-Air-Vehicles With a Wing-Stroke Dihedral Steer Passively Into Wind-Gusts. Frontiers in Robotics and AI, 9, 820363. DOI:10.3389/frobt.2022.820363. https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2022.820363/full
- Franceschini, N. et al. (2007). A Bio-Inspired Flying Robot Sheds Light on Insect Piloting Abilities. Current Biology, 17, 329-335. DOI:10.1016/j.cub.2006.12.032. https://www.cell.com/current-biology/fulltext/S0960-9822(06)02666-2
- Prasath, S. G. et al. (2022). Dynamics of cooperative excavation in ant and robot collectives. eLife. DOI:10.7554/eLife.79638. https://elifesciences.org/articles/79638
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.