Why Shape Matters: Do Our Future AI Companions Really Need to Look Human?

By Anthrobotics.in | June 2026

I. The Question That Haunts Every Robotics Lab

There is a moment that every robotics engineer remembers. It happens not at the drafting table or in simulation software, but in the quiet hours of testing — when you watch your creation try to climb its first staircase, or reach for a door handle, or simply navigate the gap between an elevator and the floor.

In that moment, a question surfaces that is at once deeply technical and profoundly philosophical: What shape should this thing be?

For decades, the answer seemed self-evident. We are building machines to live alongside us, to work in our factories, to tend to our elders, to share our homes. What shape could be more natural than our own? Two arms, two legs, a torso, a head. The humanoid form. And yet, as the robotics industry barrels toward what Morgan Stanley estimates will be a $5 trillion market by 2050, a fierce and fascinating debate is unfolding in engineering labs, boardrooms, and academic journals from Tokyo to Boston to Shenzhen.

It is not a debate about whether robots can be built. The technology has arrived. Boston Dynamics' fully electric Atlas — with its 56 degrees of freedom, 360-degree joint rotation, and 110-pound lifting capacity — entered production in 2026, its entire output already sold out. Tesla's Optimus Gen 3 has surpassed 50,000 cumulative units. Figure AI has deployed over 10,000 robots across BMW and Amazon facilities.

The question is no longer can we but should we — and the answer, it turns out, is far more nuanced than any single shape.

II. The Engineering Crucible: Legs vs. Wheels

The Bipedal Bet

Let us begin with the most visible and visceral element of the humanoid form: the legs.

A bipedal robot is an engineering marvel of compression and contradiction. It must balance a multi-kilogram mass on two narrow points of contact, constantly making micro-adjustments against the unrelenting pull of gravity. It must walk, climb, turn, crouch, stand, and recover from stumbles — all while carrying payloads and performing tasks. The computational and mechanical complexity is staggering.

Consider the numbers. According to a September 2025 report from China's Yicai Global, bipedal robots consume 40 percent of their energy and 60 percent of their computing power simply to move their legs — and in a factory setting, this locomotion generates zero productive value. Step Electric's general manager Liu Changwen put it bluntly: "I personally feel that a consensus is forming, which is that two-legged humanoid robots are unsuitable for factories."

Yet the same report reveals a countervailing truth. Puzhi Robotics, a joint venture between the Institute of PI Robotics and startup Agibot, has shipped 200 robots — 60 percent wheeled, 40 percent legged — and their bipedal units, priced between CNY 500,000 and CNY 600,000 ($70,200–$84,300), serve a distinct purpose. They handle uneven surfaces. They climb slopes and stairs. They perform in environments where wheels simply cannot go.

As PI Robotics chairman Zhou Xingyou observed: "Bipedal robots have the edge when it comes to uneven surfaces and climbing slopes and stairs. But these advantages do not matter much inside a factory."

The tension is clear. Inside a factory — flat floors, predictable routes, controlled lighting — legs are expensive over-engineering. Outside a factory — homes, hospitals, construction sites, disaster zones — legs are the only option.

The Wheeled Counter-Argument

The wheel is one of humanity's oldest inventions for a reason. It is energy-efficient, mechanically simple, and extraordinarily reliable on prepared surfaces. In the robotics world, wheeled platforms offer compelling advantages:

• Lower cost: Puzhi's wheeled models sell for CNY 400,000–450,000 ($56,200–$63,200), roughly 25 percent cheaper than their bipedal counterparts.
• Higher payload capacity: Without the weight and complexity of leg actuators, wheeled bases can carry heavier loads.
• Extended battery life: No energy wasted on balance maintenance means longer operational hours.
• Simpler control systems: Fewer degrees of freedom to manage, less sensor data to process.

Companies like Dexmate have taken this philosophy to its logical extreme, pairing wheeled bases with foldable chassis that allow floor-level manipulation and extended reach. The result is a robot that can operate for longer, carry more, and fail less often — at a price point accessible to more businesses.

The Great Compromise

The most honest answer emerging from the industry in mid-2026 is neither pure legs nor pure wheels, but a pragmatic "it depends."

A comprehensive report from The Generalist Venture Partners (TGVP) captured this perfectly: "The debate shouldn't be about which form is 'better,' but which is 'right' for the job. We advocate for a function-first philosophy: if wheels are enough, legs are unnecessary overhead. But if the terrain demands it, we shouldn't hesitate to build robots with three legs, four legs, or even eight like an octopus. The ultimate metric of a robot's success is its ability to complete tasks efficiently."

Yet there is a deeper argument — one that transcends immediate utility — that pulls the industry back toward the humanoid form. It is the argument of infrastructure.

III. The World Is Already Human-Shaped

This is perhaps the most compelling, and least discussed, argument for the humanoid form factor.

Our entire built environment — every door, every handle, every stair, every corridor, every tool, every vehicle, every workstation — is designed around the dimensions and capabilities of the human body. Door handles are at approximately one meter. Stairs have risers of roughly 18 centimeters. Light switches are at 1.2 meters. The average hallway width accommodates two people walking side by side. The average shelf height assumes a standing reach.

A robot that shares human proportions can, in theory, move through this world without a single modification. It can open any door, climb any stair, operate any tool, sit in any chair, drive any vehicle. This is not a small advantage. It is a transformative one.

As a panel at the Robotics Summit & Expo in June 2026 concluded: "The world is built for humans. If you want a robot to work in that world without rebuilding everything around it, the humanoid form starts to make a lot of sense — even if it's mechanically harder to pull off."

Consider the alternative. A wheeled robot — no matter how efficient — encounters its first staircase and stops. It finds a cable on the floor and hesitates. It approaches a door handle at the wrong height and cannot reach. These are not edge cases; they are the fundamental texture of human space. The TGVP report described it as the "last inch" problem: "A stray cable, a steep ramp, or a gap in the elevator floor can paralyze a wheeled machine. Legs treat these obstacles as stepping stones; wheels treat them as barriers."

This is why, despite the higher cost and complexity, companies like Boston Dynamics, Tesla, Figure AI, Agility Robotics, and Unitree are all betting on bipedal platforms. They are not betting against wheels. They are betting that the most versatile machine for a human-shaped world is, ultimately, a human-shaped machine.

IV. Beyond Utility: The Deep Psychology of Form

But the debate over form factor is not merely an engineering problem. It is also a deeply human one — and it reaches into the recesses of our psychology in ways that scientists are only beginning to understand.

The Uncanny Valley Revisited

In 1970, Japanese robotics professor Masahiro Mori published a brief essay that would become one of the most influential — and most misunderstood — concepts in robotics. He proposed that as robots become more human-like, our emotional response to them grows more positive — until a point. At a certain threshold of near-perfect human resemblance, the response suddenly plunges into discomfort, eeriness, even revulsion. This is the uncanny valley.

Mori's original graph, hand-drawn with pen on paper, showed affinity rising steadily as a robot moves from industrial (no human resemblance) to humanoid (some resemblance), then crashing sharply into the valley as it approaches but fails to achieve perfect human likeness, and finally rising again toward genuine human-to-human empathy.

In the decades since, research has both validated and complicated Mori's hypothesis. A 2022 meta-analysis of 488 papers on the uncanny valley, encompassing 247 effects across 49 studies (total N = 3,556), confirmed that the phenomenon has a large effect size. It is real. It is measurable. And it matters deeply for how we design robots.

But recent research has revealed something more fascinating. A 2022 study published in the International Journal of Human-Computer Studies examined emotional responses to 251 real-world robots and discovered not one uncanny valley, but two:

• One associated with highly human-like robots (Mori's original valley)
• Another associated with moderately low human-like robots

These valleys corresponded to specific perceptual mismatches — particularly between surface features (skin, hair, eyes) and body-manipulator dimensions (how the robot moves and acts). The study concluded that the uncanny valley is not a single phenomenon but a family of effects, each triggered by different kinds of mismatch between what we see and what we expect.

A 2025 paper by Karl F. MacDorman, published in the journal Perception, deepened this understanding by connecting Mori's work to Japanese aesthetic traditions spanning four centuries. MacDorman writes: "Mori's essay describes how our affinity for robots varies with their human likeness. Industrial robots, designed only to perform specific functions, do not resemble the factory workers they replace, and most people feel little affinity for them. But Mori observes that this goal is perilous. As the robot approaches but fails to become indistinguishable from human, our emotional response abruptly turns from affinity into revulsion."

The implications for design are profound. If the goal is maximum emotional affinity, aiming for perfect human likeness is a trap. As MacDarman notes: "Robots designed with surface realism — attempting to mimic human appearance exactly — can fall into the uncanny valley due to their lack of perceived authenticity. Instead, robots that focus on conceptual realism, with symbolic or stylized features, can evoke stronger emotional connections without risking the eeriness of accurate but soulless representations."

This brings us to a radical conclusion: the most humanlike robot may not be the most human-shaped one.

What the Japanese Tradition Teaches Us About Form and Spirit

Japan offers perhaps the richest cultural framework for understanding human-robot relationships. The Japanese robotics tradition has long been concerned not merely with what robots do, but with what they are — and how they relate to us.

This tradition draws on Shinto animism, in which spirits (kami) can inhabit natural objects, crafted artifacts, and even abstract concepts. A robot is not merely a machine; it can be a vessel for something more. This cultural inheritance has produced a unique approach to robot design — one that prioritizes relationship over function, and presence over performance.

Consider the concept of "weak robots" (yowai robotto), pioneered by Michio Okada at Toyohashi University of Technology. Okada's robots are deliberately designed with physical and cognitive weaknesses — they stumble over words, they move uncertainly, they need help. Why? Because weakness, in Okada's philosophy, is what makes something relatable. A perfect robot is intimidating. An imperfect one invites care.

This philosophy has been commercialized in robots like Panasonic's NiCoBo — a small, round pillow with digital eyes that communicates in "mumbling-cuddling language" (mokogo). It is the opposite of a bipedal humanoid. It makes no pretense of walking or working. And yet, studies show it forms deep emotional bonds with its human companions.

Similarly, Sony's aibo — a robotic dog with no pretension to human form — has been shown in a 2026 study published in Intelligent Human Systems Integration to generate "a strong sense of emotional reciprocity" in its owners. The study found that many owners perceive aibo as "an entity that 'needs them' and 'responds to their care,'" contributing to "purpose, enjoyment, and everyday emotional stability."

The Japanese approach suggests that shape is not destiny. A wheeled companion robot, a pillow-shaped care robot, or even a four-legged robotic pet can forge bonds as deep as any bipedal humanoid — precisely because they do not trigger the perceptual mismatches that cause the uncanny valley.

V. The Market Speaks: Who's Building What, and Why

As of mid-2026, the humanoid robotics market has fractured into distinct form-factor strategies, each backed by serious capital and real-world deployment.

The Bipedal Believers

Tesla Optimus Gen 3 — Over 50,000 units produced. Deployed in Tesla's own Gigafactories (Austin, Shanghai, Berlin) performing pick-and-place, battery cell sorting, and light assembly. Price target: under $20,000. The bet: mass production at consumer-electronics scale will make legs affordable.

Boston Dynamics Atlas (Electric) — The gold standard of bipedal mobility. 56 degrees of freedom, 360-degree joint rotation, 110-pound lifting capacity. 2,000–5,000 units in limited production. Entire 2026 output sold out to Hyundai and Google DeepMind. Lease pricing: $150,000–$250,000/year. The bet: unmatched mobility justifies premium pricing.

Figure AI (Figure 02) — 10,000+ units deployed at BMW and Amazon facilities. OpenAI-powered conversational interface. Price: $50,000–$70,000 per unit. The bet: legs combined with advanced AI create a general-purpose labor platform.

Agility Robotics (Digit) — 3,000+ units in warehouse logistics. Lease pricing: $100,000–$150,000/year. The bet: purpose-built bipedal design optimized for last-meter delivery.

Unitree (H1/G1) — 5,000+ units sold. The G1, at $16,000–$20,000, is the most affordable advanced bipedal robot on the market. Targets education, research, and light commercial use. The bet: democratizing access to bipedal platforms.

The Pragmatists of Form

1X Technologies (NEO) — A general-purpose humanoid designed for the home. First customer deliveries began in 2026. Features wheeled locomotion for most indoor environments but maintains human-like proportions for manipulation and interaction. The bet: hybrid form factors that match form to function.

Dexmate — A wheeled humanoid with a foldable chassis. Higher energy efficiency, longer battery life, lower cost. The bet: most real-world tasks don't require legs.

Real Bobox (Melody / David) — Companion robots with modular, customizable faces and personalities. Move on wheeled bases, not legs. Development priorities center on "social and emotional intelligence rather than locomotion." The bet: for companionship, personality matters more than legs.

What the Data Tells Us

The 2026 landscape reveals a clear pattern: the form factor market is segmenting by use case.

• Industrial/logistics (factories, warehouses): Leaning toward legs, but with significant wheeled alternatives
• Healthcare/elder care: Hybrid approaches, with form factor determined by specific care tasks
• Home/companion: Wheeled bases and non-humanoid forms dominate, prioritizing emotional connection over mobility
• Research/education: Full bipedal platforms for experimentation and skill development
• Specialized environments (construction, disaster response): Legs are non-negotiable

The data from China is especially instructive. At the 2025 Apsara Conference, Puzhi Robotics reported that 60 percent of their shipped units were wheeled and only 40 percent were legged — and the legged units were "mainly used for exhibition hall tours and demonstrations" rather than actual work. Yet Puzhi's chief scientist Jiang Lei countered: "The positioning accuracy of wheeled robots is only at the centimeter level. But industrial applications require sub-millimeter and millimeter precision. This is very hard to achieve with a wheeled base as it cannot provide a perfectly stable reference point."

It is this tension between practical deployment today and theoretical capability tomorrow that defines the current moment.

VI. The Humanistic Dimension: What We Lose and Gain by Shape

Beyond the engineering and the markets lies a question that is ultimately humanistic in the deepest sense.

When we build a robot in our image, what are we saying about ourselves? And when we choose a different shape — a wheeled base, a pillow-like companion, a robotic pet — what do we gain?

The Risk of Anthropocentrism

There is a danger in designing robots that look too human. It is not merely the uncanny valley. It is the risk of creating expectations that machines cannot fulfill. A human-shaped robot that cannot hold a conversation, cannot understand emotion, cannot truly care — this is not just a technical failure. It is an existential disappointment.

The philosopher and roboticist Hiroshi Ishiguro, who has built androids in his own image, has spent decades exploring this boundary. His work reveals that when a robot looks sufficiently human, people begin to treat it as human — they expect reciprocity, empathy, understanding. When those expectations are not met, the result is not neutral. It is alienating.

This is the paradox at the heart of the form-factor debate. The more human a robot looks, the more we demand of it — and the more it disappoints us by being merely a machine.

The Liberation of Non-Human Forms

By contrast, robots that do not pretend to be human can excel at being companionable machines. Sony's aibo is not a failed dog. Qoobo — a haptic creature by Yukai Engineering that is literally a pillow with a wagging tail — is not a failed pet. LOVOT by Groove X is not a failed child. They are successful robots precisely because they set appropriate expectations.

Research published in the 2026 volume Intelligent Human Systems Integration found that these haptic creatures, with their rounded designs, soft materials, and limited but expressive movements, generate deep emotional bonds precisely because they do not trigger the perceptual mismatches that cause the uncanny valley. They invite touch. They invite care. They invite relationship — on terms that a machine can actually fulfill.

The Japanese concept of "techno-intimacy" — intimacy formed between humans and machines — offers a framework for understanding this. As scholars have documented in recent ethnographic studies of Japanese companion robot culture, the most successful robot companions are not those that simulate human form most accurately, but those that create conditions for affective presence — a sense of being with, not just being used.

VII. Toward a Philosophy of Form

So where does this leave us? Do our future AI companions really need to look human?

The answer, emerging from the convergence of engineering data, market evidence, psychological research, and cross-cultural philosophy, is more nuanced than a simple yes or no.

Yes, they need the human form when they must navigate a human-built world — climbing stairs, opening doors, using tools, entering spaces designed for bodies like ours. This is the domain of the bipedal work robot — the Atlas, the Optimus, the Figure 02 — and for these applications, legs are not optional. They are the price of admission.

No, they do not need the human form when their primary function is relationship — companionship, care, emotional presence. In these domains, non-human forms often perform better precisely because they do not raise expectations that cannot be met. The pillow-like NiCoBo, the tail-wagging Qoobo, the cuddly PARO — these are not failures of human form. They are successes of appropriate form.

Maybe, they need an intermediate form — a human-like torso on a wheeled base, or a bipedal platform designed for work but not for intimacy. This is the pragmatic middle ground that many companies are exploring, and it may prove to be the most commercially viable approach in the near term.

The Shape of Things to Come

What is clear is that the era of a single dominant form factor is over before it began. The future is not one robot shape, but many — each matched to its purpose, its environment, and its relationship with humans.

The TGVP report concluded with a statement that captures the emerging consensus: "We must define robots by their utility, not limit them to a specific shape."

And yet, this pragmatic conclusion does not fully capture the humanistic heart of the question. For utility is not the only measure. There is also meaning. There is connection. There is the strange and wonderful possibility that by building machines that are different from us, we might learn something new about ourselves — that otherness can be as valuable as sameness.

Mori himself, in the final paragraphs of his 1970 essay, offered a vision that transcends engineering. He wrote that the goal of robotics should not be to create perfect copies of humans, but to create beings that "make us feel more human."

Perhaps that is the deepest argument for diversity of form. A world of identical human-shaped robots would be a hall of mirrors, reflecting only ourselves. A world of robots in many shapes — some bipedal, some wheeled, some soft, some hard, some that walk and some that roll and some that simply sit beside us and keep us company — such a world would be richer, more surprising, more alive.

That, in the end, may be what shape really matters for.

References

1. Mori, M. (1970). The Uncanny Valley. Energy, 7(4), 33–35. (Translated by K. F. MacDorman & T. Minato, 2005).
2. Mori, M., MacDorman, K. F., & Kageki, N. (2012). The Uncanny Valley [From the Field]. IEEE Robotics & Automation Magazine, 19(2), 98–100.
3. Diel, A., Weigelt, S., & MacDorman, K. F. (2022). A Meta-Analysis of the Uncanny Valley's Independent and Dependent Variables. ACM Transactions on Human-Robot Interaction, 11(1), 1–35.
4. Mara, M., Appel, M., & Gnambs, T. (2022). Human-Like Robots and the Uncanny Valley: A Meta-Analysis of the Curvilinear Relationship Between Anthropomorphism and Likeability. University of Würzburg.
5. MacDorman, K. F. (2025). Chikamatsu, Mori, and the Uncanny Valley. Perception, 54(3), 1–28.
6. The Generalist Venture Partners (TGVP). (2026). Humanoid Robots in 2025: TGVP Report.
7. Bain & Company. (2025). Humanoid Robots: From Demos to Deployment — Technology Report 2025.
8. Yicai Global. (2025). Wheeled Robots Gain Edge Over Bipedal Peers in China's Factories. September 29, 2025.
9. RobotDevDiary. (2026). Humanoid Robot Design in 2026: Legs, Looks, and Leadership. June 14, 2026.
10. Boston Dynamics. (2026). Atlas Humanoid Robot — Product Specifications.
11. AIMagicX. (2026). Humanoid Robots in the Workplace: The 2026 Business Leader's Guide. March 23, 2026.
12. Humanoid Robotics Technology. (2026). Top 12 Humanoid Robots of 2026. June 24, 2026.
13. Okada, M. (2012, 2017). Weak Robot Theory and Human-Machine Bonding. Toyohashi University of Technology.
14. Aronsson, A. S., & Holm, N. (2022). Theorizing the Real in Social Robot Care Technologies in Japan. International Journal of Japanese Sociology.
15. White, D., & Katsuno, H. (2021). Tactile Affect and Human-Robot Intimacy in Japan. In The Future of Humans and Emotional Machines. OAPEN Library.
16. Breazeal, C. (2003). Emotion and Sociable Humanoid Robots. International Journal of Human-Computer Studies, 59(1–2), 119–155.
17. Intelligent Human Systems Integration (IHSI). (2026). Emotional Engagement and Well-Being in Companion Robot Ownership in Japan. Vol. 200, pp. 706–712.
18. MacDorman, K. F., & Chattopadhyay, D. (2016). Reducing Consistency in Human Realism Increases the Uncanny Valley Effect. International Journal of Human-Computer Studies, 96, 27–38.
19. Yam, K. C., et al. (2024). Rapid Processing of Uncanny Stimuli: 50-ms Exposure to Humanlike Robots. Journal of Experimental Social Psychology.
20. Morgan Stanley Research. (2025). Humanoid Robot Market Projections: $5 Trillion by 2050.

This article was written for Anthrobotics.in — exploring the humanistic frontier of artificial intelligence and robotics. For more deep-dive analyses on the future of human-robot coexistence, subscribe to our newsletter.