SE01's Gait Is Real, But Don't Rush to Talk About It Doing Work

A friend recommended EngineAI's SE01 to me. His exact words were, you've definitely seen that video of it strolling around the streets of Shenzhen, come check out the real thing at their showroom. I was on a business trip to Shenzhen last week and freed up half a day.

Let me be clear upfront: you can't take this thing home as an individual. It's 170 cm tall, 55 kg, a bit taller than me. My friend's company does showroom services and got one to draw foot traffic. I saw the real machine in their showroom, spent half a day there, and even operated it myself a few times with the teach pendant. The teach pendant is that handheld remote control box, pretty much the same setup as industrial robot arms.

A bit of background. I'd only seen humanoid robots from afar at expos before; my main work is in intelligent driving. The SE01 is EngineAI's first full-size humanoid — "full-size" meaning built to human height and weight, not a desktop toy. It has 32 degrees of freedom total; you can think of a degree of freedom as a joint that can rotate independently — the higher the number, the finer the motion. Official walking speed is 2 meters per second.

The first time I watched it walk, honestly, it felt a bit off. With traditional humanoids you can tell at a glance it's a machine — knees slightly bent, tiny shuffling steps, like walking on stilts. The SE01 isn't like that. It lowers its center of gravity a bit before stepping, then swings the leg, heel strikes and rolls to the toe, and the rhythm is about the same as how I normally walk. That 2 m/s figure, seeing it live, is basically an adult's brisk walking pace.

After Nvidia's Jim Fan saw that Shenzhen street footage, his first reaction was to ask whether it was real or Sora-generated. After seeing the real machine, I can kind of understand why he asked that.

The perception setup is pretty generous. 360-degree lidar, an Intel RealSense D435 depth camera, 6 HD cameras, plus infrared. This logic is in the same lineage as intelligent driving — lidar produces the point cloud, cameras produce texture, infrared fills in for low light, and after fusion it does obstacle avoidance and path planning. My friend set up three cardboard boxes in the showroom and it went around them on its own. The path wasn't optimal, but it didn't hit anything.

Now for where it gets stuck.

One: the end-to-end neural network is a black box. That's the core of the official pitch — letting the robot say goodbye to mechanical gait. The problem is a black box is painful to tune. Their engineer complained to me that when you change a parameter for the same motion, sometimes it works and sometimes it collapses, so you just have to try again and again. Same flaw as end-to-end in intelligent driving — when a corner case hits, it's hard to say which layer caused the problem.

Another: the upper body can do limited work. Most of the 32 degrees of freedom go to the legs and torso; fine manipulation with the hands isn't a strength. On site it handed me a water bottle — slow motion, and the grab position was set up by me. If you're expecting it to go onto a production line and tighten screws or move material bins, that's not happening right now.

I didn't get exact data on battery life and heat dissipation. My friend said continuous demos have to be timed with a stopwatch, with buffer time left for the joints in between. I can't verify that claim, but joint heating is a common problem for humanoids, so it's not surprising.

On price, media reports put the final retail price at around 150k RMB. Not cheap for an individual, but for a small showroom it's an investment you can actually calculate.

So the conclusion is: it depends — not a simple recommend or don't recommend.

Who it's for: showroom traffic, brand events, science museums, universities doing robot algorithm research. What these scenarios want is precisely the fact that it can walk like a human, and the SE01 does that on a real machine, not in an edited clip. From what I saw, the gait is genuinely solid.

Who it's not for: factories that want to buy it to replace human labor, and individuals who want to buy it as a home helper — stay away from both. Its current positioning is display and platform, not labor.

Looking ahead, I'd bet that within a year and a half, humanoid robots won't be competing on whether they can walk naturally anymore — walking will become the entry ticket. The real watershed will be upper-body manipulation precision and continuous-operation reliability. The SE01 has turned in its paper on the gait test; the next test has only just begun.