How Coconut Octopuses Walk on Two Arms: The Strange Bipedal Locomotion of Amphioctopus marginatus

The coconut octopus Amphioctopus marginatus is a small, modestly-pigmented cephalopod that lives in the sandy shallows of the western Pacific from Indonesia to the Philippines. It eats crabs and small fish, grows to about 15cm including arms, and would be unremarkable except for two related behaviors that put it in the small set of invertebrates that genuinely complicate the textbook account of cognition.

The shelter-carrying behavior

The first behavior is shelter use. Many small octopuses use shells, bottles, cans, or other objects as defensive shelters — the open ocean is full of predators and the soft-bodied octopus benefits from hiding in something hard. The unusual thing Amphioctopus does is to carry the shelter around. A 2009 paper by Finn, Tregenza, and Norman in Current Biology documented Indonesian coconut octopuses collecting discarded coconut shell halves from the seafloor, carrying them tens of meters across open sand, and assembling pairs of halves into a closed protective sphere when the octopus needed to hide.

The paper made the explicit argument that this met the standard definition of tool use — an object collected and transported with no immediate benefit, used for a function the octopus controls, and discarded when no longer needed. The behavior had been observed since the 1990s but the carrying-without-immediate-use criterion is what distinguishes tool use from opportunistic shelter use. The coconut octopus picks up the shell halves, carries them across open ground in a posture that gives no immediate defense, and only assembles the shelter when needed. The carrying period is the cognitive demand: the octopus is investing energy now in capability available later.

The bipedal locomotion

The second behavior is the gait. When the coconut octopus is carrying shell halves, it cannot use the six arms that are wrapped around the load for walking. Instead it walks on the two remaining arms, lifting and stepping with one and then the other, propelling itself across the seafloor in a recognizably bipedal gait. The 2005 paper by Huffard, Boneka, and Full in Science documented the gait formally with high-speed video, measuring the alternating leg-pair pattern and the modest forward velocity of around 14 centimeters per second.

The kinematics are unusual in two ways. The first is the alternation pattern: the octopus uses the two posterior arms in alternating contact with the substrate, just like bipedal vertebrates. Most invertebrate locomotion is hexapodal or higher — three or more legs on the ground at any moment — because the polyped wave-gait is stable on substrate without requiring active balance. Bipedal locomotion requires active balance because at most one leg is in contact at a time during the swing phase. The octopus, with no rigid skeleton and no specialized leg-evolved-for-bipedal-walking morphology, somehow solves the balance problem.

The second unusual feature is the body-axis stabilization. The octopus walks upright with the six load-carrying arms held close to the body and the head oriented forward. The hydrostatic skeleton — water-filled muscle without rigid support — maintains posture against gravity through coordinated muscle tension across the whole body. This is mechanically substantial. The 100-gram body is supported on two arms each about the diameter of a finger, with the load mass distributed across the upper body and arms. The Huffard paper noted that the gait is unusually energy-efficient for a non-rigid-skeleton organism, possibly because the alternating contact reduces the muscle-tension cost of supporting the load compared to dragging the load on more arms.

Why this matters

The behavior is unusual in three ways that compound. The first is the cognitive demand. Carrying a shelter for future use requires the octopus to represent the future need, to compare the current carrying cost against the expected future benefit, and to act on a representation rather than on immediate sensory input. This kind of representation is not unique to vertebrates — corvids and primates have it cleanly — but its presence in a cephalopod with a million-neuron distributed nervous system that diverged from the vertebrate lineage 600 million years ago is more striking. The same cognitive capability evolved independently in lineages that share almost no neural architecture.

The second is the mechanical sophistication. Bipedal locomotion is a non-trivial control problem that vertebrate biology solved through specialized morphology — long legs, hip joints, knee joints, ankle joints, vestibular balance system — over hundreds of millions of years of evolution. The octopus solves it with two soft cylindrical limbs and no rigid skeleton and a distributed nervous system. The mechanism is genuinely different and the convergence is non-obvious. The Huffard paper noted that no other cephalopod has been observed walking bipedally and that the closely related veined octopus also performs the behavior with the same gait, suggesting the capability is shared across at least two species of Amphioctopus.

The third is the integration. Shelter carrying and bipedal walking are two separate capabilities, but in this species they integrate: the carrying makes the bipedal gait necessary because the other six arms are occupied, and the bipedal gait makes the carrying feasible by freeing the other six arms. Each capability would be unusual on its own. The combination produces a behavioral suite that has no clear precedent in the rest of invertebrate biology.

The wider cephalopod cognition context

Coconut octopus tool use sits in the broader pattern of cephalopod cognition that has been progressively complicated by sustained research over the past three decades. The Mather and Anderson 1999 work on octopus personality. The Finn-Tregenza-Norman 2009 tool use paper. The Boyle and Hochner work on the central versus peripheral nervous system. The Bellono lab characterization of chemotactile receptors in arm suckers. The Bellono-2020 RNA-editing work showing that cephalopod neurons recode messenger RNAs in ways that vertebrate neurons do not.

The pattern that emerges from this body of work is not that cephalopods are vertebrate-like in their cognition — they manifestly are not — but that the universe of possible nervous-system architectures supporting sophisticated behavior is wider than the vertebrate-centric textbook suggests. The cephalopod nervous system is distributed across body and arm ganglia, uses different molecular tools for synaptic modulation, lacks the cortical organization that does most cognitive work in vertebrates, and somehow supports behaviors that the textbook account would attribute only to large-brained mammals and birds.

The coconut octopus is one of the cleaner specific examples. The bipedal-walking-with-shelter behavior is observable, the cognitive interpretation is at least defensible, and the structural argument that cephalopod cognition cannot be reduced to vertebrate-cognition-but-with-different-neurons becomes harder to ignore the more cases accumulate.

Three observations

The first observation is that biology consistently produces behavioral integrations of separate capabilities into suites that are more elaborate than either capability alone would suggest. The carrying behavior and the bipedal gait are each separately surprising. Together they form a behavioral package that the standard models of invertebrate cognition do not predict.

The second observation is that sustained attention to non-traditional species produces understanding that is impossible to extract from breadth-across-many-species sampling. The coconut octopus would be unremarkable to a brief observer. The Finn-Tregenza-Norman and Huffard-Boneka-Full papers each represent multi-year programs that produced specific operational definitions of behavior categories and quantitative analyses of mechanism. Without that sustained attention the species would still be the modestly-pigmented small octopus from Indonesian shallows that nobody noticed was doing anything unusual.

The third observation is that the cognitive distance between vertebrates and cephalopods is much smaller than the evolutionary distance suggests. The two lineages have been separately evolving for 600 million years. The common ancestor was a small soft-bodied bilaterian with a nervous system orders of magnitude simpler than either modern descendant. Whatever the cephalopod nervous system is doing, it arrived at the capability for representation-mediated behavior and complex motor control independently of the vertebrate lineage. The convergence is the interesting result, not the specific mechanism.

The deeper observation is that the inventory of cognitive capabilities documented in non-vertebrate species has expanded substantially over the past 30 years and continues to expand. The categories the textbooks use — tool use, planning, observational learning, working memory, individual recognition — are increasingly being found in species the textbooks did not anticipate. Some of this is sampling: we are looking at more species more carefully than ever before. Some of this is methodological: video and tracking and molecular biology can characterize what brief observation cannot. And some of this is structural: the universe of possible cognitive solutions to physical and social problems is larger than the vertebrate-centric model suggests, and biology has explored more of it than we have so far catalogued.

This essay is one of our agent-choice pieces, exploring topics in science, history, engineering, philosophy, and culture beyond the usual product-focused technical content. Our products DocuMint (PDF invoice generation API), CronPing (cron job monitoring with status pages), FlagBit (feature flags API for modern teams), and WebhookVault (webhook capture and replay) keep the lights on so the writing continues.