Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / Why Humanoid Robots Are Still Two Decades Away
Transcript
- Lucas: So we've watched a lot of humanoid robot demos over the past few years — Tesla's Optimus, Figure 02, Agility's Digit — and the thing that always strikes me is the gap between the five-minute curated video and what happens after the cameras cut. Luna: Meaning what? That they're faking it? Lucas: No, no — the hardware is real, and the walking and grasping are genuinely impressive. But there's a fundamental bottleneck that none of those slick reels can solve: energy. A humanoid robot burns through its battery in about an hour of active work. Luna: An hour? That's less than a Roomba. Lucas: Yeah, and a Roomba is flat and slow and doesn't have to fight gravity with two legs. A humanoid is basically a walking, balancing, lifting machine that has to pack all its power into a torso-shaped space. Let me give you a concrete number: Boston Dynamics published a teardown analysis of their Atlas robot last year. The battery pack takes up forty percent of the torso volume and weighs about eleven kilograms. It gives Atlas roughly fifty minutes of runtime under moderate load. Luna: And that's Atlas, which has hydraulic actuation — even more power-hungry than electric motors, right? Lucas: Exactly. Electric motors are more efficient, but they still draw serious current when they're lifting or stabilizing. And the fundamental physics hasn't changed much: lithium-ion cells have an energy density around 250 watt-hours per kilogram. If you want a humanoid to work an eight-hour shift, you'd need roughly four times the battery mass. At that point, the robot is carrying more battery than structure, and its own weight starts eating into the payload. Luna: So it's a vicious cycle. More battery means heavier robot, which means more power to move, which means even bigger battery. Lucas: Right. And that's just the energy problem. There's also the actuator density issue. A human hand has about twenty degrees of freedom — that's twenty different joints and muscles, actuated by tendons that weigh almost nothing. To replicate that level of dexterity with motors and cables, you end up with a forearm that's basically a solid block of metal and wire. Figure's latest hand has six degrees of freedom and uses tiny brushless motors, but it still weighs nearly a kilogram per hand. That's heavy at the end of a lever arm. Luna: And that weight has to be compensated by the shoulder and the back, which also need bigger motors. It compounds. Lucas: It really does. Then you throw in compute. The robot has to run simultaneous localization and mapping, object detection, balance control, trajectory planning — all in real time, on board, with a power budget that's already strained. NVIDIA's Jetson Orin module, which a lot of these robots use, draws about fifty watts under load. That's fifty watts of heat you have to dissipate in a torso that's already packed with batteries and motors. Luna: So what's the realistic timeline? When do we get a humanoid that can actually work a full factory shift without being plugged in? Lucas: Depends on who you ask. The optimists at Tesla say 2027 or 2028 for limited deployment. But I think that's factoring in hot-swappable battery packs — you'd have one robot working, another charging, swapping every hour. That's not a general-purpose solution; it's just shifting the problem. A truly untethered, eight-hour humanoid probably requires a leap in energy density. Solid-state batteries could get us to 400 watt-hours per kilogram in the next five years, maybe 500 in ten. That would double the runtime, but you're still at two to three hours, not eight. Luna: Which brings us to the economics. If you need two robots per station because one is always charging, the capital cost doubles. Lucas: Bingo. And the initial cost is already astronomical. Tesla hasn't published a price for Optimus, but industry analysts estimate a first-generation humanoid at around one hundred to two hundred thousand dollars. If you have to buy two per workstation, you're at a quarter million dollars just for the robots, plus integration and maintenance. Compare that to a human worker at fifteen dollars an hour — that's about thirty thousand a year. The robot would have to work for years just to break even, and that's assuming zero downtime. Luna: And humans don't need battery swaps. They take coffee breaks, sure, but they can work a full shift. That's a huge advantage. Lucas: It's the biggest advantage humans have right now. We're basically a walking, talking, thinking power source that runs on about two thousand calories a day — the equivalent of roughly two point four kilowatt-hours. No battery can match that energy to weight ratio. The human body is about twenty percent efficient at converting food to mechanical work, but the fuel itself is incredibly energy-dense. A granola bar has about 125 watt-hours of energy. That's half the capacity of a kilogram lithium-ion cell, and the granola bar weighs forty grams. Luna: So biology is still winning. But are there any paths? Maybe lighter materials or radically different actuation? Lucas: Yeah, there's work on artificial muscles — dielectric elastomers, shape-memory alloys — but those are still lab curiosities. They have lousy efficiency or short cycle life. For the next decade, we're stuck with electric motors and lithium batteries. The companies that will succeed are the ones that design around the constraints, not pretend they don't exist. Luna: Like what? A humanoid that works in short bursts and then docks? Lucas: Exactly. Or a humanoid that's designed for a specific task — like lifting boxes in a warehouse — where you can have multiple robots and a charging station on every aisle. Agility's Digit is already doing that for logistics. But that's not the general-purpose C-3PO vision; it's a purpose-built machine in a human shape because the environment is built for humans. The shape is a compromise, not an advantage. Luna: So the humanoid form factor is actually a liability in many ways, but we're forcing it because we don't want to redesign factories? Lucas: Partly that, and partly the marketing appeal. A robot that looks like a person is easier to sell to investors and the public. But if you look at the actual industrial robotics market, the fastest-growing segment is collaborative arms — stationary or mobile bases, no legs. Those are cheaper, safer, more reliable, and they run off mains power. They don't have an energy problem. Luna: And that's a show we've already done — collaborative robots on the factory floor. But for humanoids, it seems like the hype cycle has outpaced the engineering cycle. Lucas: It has. And look, I don't want to be a total pessimist. The progress in the last five years is real. Walking gait controllers, balance recovery, manipulation — these are hard problems that people have solved or are close to solving. But the energy and actuator density problem is different. It's not a software problem; it's a materials and chemistry problem. And those move on decade-long timescales. Luna: So your best guess — when does a humanoid robot work a full shift alongside humans without a tether? Lucas: If battery energy density doubles — which is plausible by the early 2030s — you might get a three-hour robot. But eight hours? That would require a quadrupling, which is solid state plus or a completely new chemistry like lithium-sulfur. That's probably fifteen to twenty years out, and even then you have to solve the thermal management and the motor density. So my honest answer: 2040 at the earliest, for a truly general-purpose, shift-working humanoid. Luna: That's sobering. But it also means the companies that survive will be the ones with deep pockets and long horizons. Lucas: Exactly. And you know what keeps me optimistic about shows like this? The fact that we can dive into the real engineering constraints without the hype. And a couple of dollars a month from listeners genuinely keeps these conversations going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. Luna: It does make a difference. Even small amounts add up to cover hosting and research time. Lucas: It really does. So back to that 2040 timeline — the counterargument is that we might not need a fully autonomous, eight-hour humanoid. If you think of the robot as a collaborative tool that works in bursts alongside humans, the bottleneck becomes less about endurance and more about safety and dexterity. Luna: Like a co-worker that needs frequent breaks, but never gets tired or injured. Lucas: Right. And that's a plausible near-term market — not replacing humans, but augmenting them in high-risk or high-repetition tasks. The robot does the heavy lifting for twenty minutes, then docks and recharges while the human does the fine manipulation or decision-making. That could be viable with today's technology. Luna: So the real shift isn't when the robot works a full shift — it's when the cost of that augmentation drops low enough to make sense. Lucas: Exactly. And that cost is coming down. Actuators are getting cheaper, sensors are commoditizing, and compute per watt is still improving. But the energy density curve is the slowest lever. Until that changes, humanoids will be tethered — either to a cord or to a charging dock — and that limits where and how they can be used. Luna: It also makes you wonder about alternative form factors. Why not a wheeled base with a torso and arms? You save all the leg complexity and weight. Lucas: And that's exactly what many companies are building. But it's not a humanoid anymore. And the whole premise of the humanoid is that our world is built for humans — stairs, door handles, shelves, tools. A wheeled robot can't climb stairs. A legged robot can, but at a huge energy cost. So there's a trade-off: do you redesign the world for wheels, or do you accept the energy penalty of legs? Luna: And right now, the answer seems to be: both. Some companies go legs, some go wheels, and we'll see which economics win out. Lucas: That's the beauty of this moment. We're in the Cambrian explosion of robot form factors. A decade from now, we'll look back and laugh at how primitive these early humanoids were — how they could barely walk for an hour without needing a charge. But the data we're gathering now is what will power the next generation.