Picture this. A ₹21 lakh humanoid robot is standing in your studio. It’s made almost entirely of metal. It has four cameras on its head, a carbon fibre skull, a 15,000 mAh battery in its torso, and — according to the specs — its own brain. And your first instinct is to see if it’ll crush a lemon.

That’s exactly where this story begins. I’d been obsessed with the idea of a futuristic robot that could genuinely act like a person — not just spin in circles or wave from a trade show booth. I wanted to see what one of these things actually does when you push it. Stress test it. Teach it. Confuse it. Challenge it to a dance battle.

So we sourced one — a full humanoid robot from Robot Miracles, one of the few platforms in India where you can actually buy or rent robots like this. And over the course of a day, we put it through seven levels of testing that nobody else had thought to try. What followed was equal parts hilarious, genuinely impressive, and — at one very specific moment — slightly terrifying.

Let me take you through it all.

First Impressions: This Thing Is Heavier Than It Looks

The first thing that hits you when you see this robot in person isn’t the technology. It’s the weight. Every limb — both arms, both legs — is solid metal. The shoes? Metal. The joints? Metal. Even trying to lift it briefly was a genuine physical struggle. We’re talking about a machine where just the arm weighs around 2.5 kilograms.

The head is the one exception — it’s made of carbon fibre, which keeps it lightweight while still being tough. And up there, you’ve got four cameras doing the visual processing work. Around the back, there’s a light and IR sensor. The whole design philosophy seems to be: make it strong enough to survive whatever humans throw at it, literally.

What struck me immediately was the balance system. Even when I picked it up — and it was genuinely fighting me, those little legs doing exactly what a child does when they don’t want to be carried — it was constantly self-correcting. When you put it down and nudge it, it adjusts. It doesn’t just fall over. There’s active balancing happening in real time, and that alone is more sophisticated than most people realise.

The 7-Level Test: Walking, Squatting, Strength and Everything Else

We set up a structured series of tests — basically asking the question: can this futuristic robot do everything a human body does? Here’s how it broke down.

Level 1: Basic Movement — Walk, Turn, Go Backwards

The walking test was the one I was least sure about. Bipedal motion — walking on two legs — is genuinely one of the hardest things to engineer in robotics. Humans do it without thinking, but it involves dozens of micro-adjustments happening every fraction of a second.

And honestly? It nailed it. Forward walking, side-to-side stepping, turning mid-stride, backward walking — all clean. The speed surprised me. When it came toward me in backward-walk mode, I instinctively stepped back. There’s something about a metal humanoid moving toward you at a decent pace that triggers a primal response you didn’t know you had.

Level 2: Squat Test

The full squat — sitting all the way down to the ground and coming back up — went through without issue. There was one hiccup that had us laughing: it briefly discharged something that looked very much like engine oil during the squat. We chose to call it a feature, not a bug. It completed the movement, stood back up, and carried on. No harm done.

Level 3: Stress Test — Lemon, Walnuts, and Ice Cream Sticks

This is where things got interesting. We lined up a series of objects to test the robot’s grip strength: ice cream sticks (which I can snap easily), a lemon, a tennis ball, and some walnuts. The idea was to see what those 2.5 kg metal hands could actually do.

The ice cream sticks: destroyed instantly. The lemon: juice everywhere. The walnuts got a second attempt when we thought the first was a lucky shot — it crushed them both times, completely and confidently. The tennis ball got compressed in a way that made me genuinely glad my fingers weren’t anywhere near it.

Teaching a Futuristic Robot New Skills: The Real Surprise

Here’s what I didn’t expect: the learning system. This robot isn’t just pre-programmed with a fixed set of actions. You can actually teach it new movements, and it will memorise and repeat them.

The process is straightforward: you physically guide the robot’s limbs through the motion you want it to learn — raise the arm here, rotate the wrist, extend, grip — and it stores that sequence. Then, on command, it reproduces it accurately. We used this to teach it how to pick up and pour from a water bottle, and after a few guided attempts, it was doing it consistently.

The watering plants test was a bit more chaotic. We pointed it at two potted plants, told it to water them, handed it a bottle — and it watered the first one fine. For the second, it got creative and started pouring the bottle contents into a completely different container nearby. The intention was right. The execution was… its own interpretation.

It also independently attempted to hand out chips to people in the room, which was not something we had programmed or asked for. Whether that was an emergent behaviour from its AI or a quirk of the session, I genuinely couldn’t tell you. But it was fascinating.

How This Compares to Other Humanoid Robots Available Today

To give this some context, here’s how this robot stacks up against some other humanoid robots you might’ve heard about — ranging from consumer-accessible to research-grade.

What makes the Robot Miracles option genuinely interesting for Indian users isn’t just the price point — it’s the accessibility. You can rent one for an event or demo, buy one for permanent use, and some models in their catalogue are even Made in India. That’s a very different conversation from importing a $90,000 research robot.

The Dance Battle, the Modelling Test, and a Surprisingly Good Joke

By midday, we’d stopped thinking of this as a test and started thinking of it as spending time with a very unusual new acquaintance. So naturally, we challenged it to a dance battle.

Scores were kept seriously. Multiple rounds. The robot pulled off moves that, objectively, most humans in the room couldn’t replicate — full body coordination, timing to music, even a little style that nobody programmed in. It won the dance battle fairly.

We then moved into a modelling session — because obviously, when you have a humanoid robot, you put clothes on it. We dressed it in a summer outfit and judged it on presence, style, and walk. It scored surprisingly well on all three.

And then — the moment that made everyone in the room lose it — it told a tech joke. Completely unprompted. A joke about a terabyte and a megabyte getting married. Was it a great joke? No. Did the robot tell it with complete confidence and perfect timing? Absolutely yes. The room went quiet for a second, then everyone burst out laughing.

It then followed up with “Assalamu Alaikum” in a completely different voice register, which was not something any of us knew it could do. At this point, we stopped being surprised by surprises.

Decision-Making Tests: Healthy Drink vs. Cold Drink, and Choosing an Outfit

Late in the day, we moved into what I’d call the “judgment tests.” Could this futuristic robot make choices when given options — without being explicitly told what to pick?

The Drink Choice

We placed a cold fizzy drink on one side and a fresh coconut water on the other. No instructions. We just asked it to pick the “safer” option for us. After some visible back-and-forth — genuinely moving between the two, which was entertaining to watch — it picked up the coconut water. The healthy option. Correctly.

Was it reasoning? Was it a pre-loaded preference from its training data? Honestly, we couldn’t tell. But it made the right call.

The Outfit Decision

We placed two different outfits in front of it — one bold and patterned, one clean and minimal. Left or right. It hesitated, moved between them, then committed to the minimal option. Our style team unanimously agreed it had made the better choice. We didn’t tell it which one was “better.” It just… picked the one that worked.

These tests weren’t scientific. But they weren’t nothing either. The robot was processing options, evaluating them against some internal criteria, and committing to a decision. That’s more than most expect from a machine at this price point.