Don't rush the ev¹ onto the road—teach it a regular route first

A friend recommended the Strutt ev¹ to me, so I tried it to see whether it's actually easy to use. He works in elderly care facility operations, and the 3–5 km short shuttle routes in the campus have always relied on caregivers pushing wheelchairs, so he asked me whether this vehicle can run by itself. I borrowed one and used it for a few days, got the most practical function working end to end, and I'm writing it down for others who are just starting out.

Background first. This vehicle is made by Ruochuang Technology, went on sale offline in China on August 29, 2026, and previously won CES and iF design awards. Founder Hong Xiaoping used to run the optoelectronics division and the lidar product line at DJI, so it's no surprise the vehicle has lidar on it.

Hong Xiaoping himself says this is a new kind of mobility device, a different thing from a traditional smart wheelchair. I didn't believe it at first either. After getting hands-on, I found the difference really isn't in the seat — it's in the sensors and the system called "Companion Smart Driving." The vehicle finds its own way and dodges people itself; you're only responsible for telling it where to go. Underneath there's also a "Lingbo chassis" whose wheels can actively adjust their posture, so going over bumps isn't jarring.

This post teaches only one thing: teaching it a route you walk every day, then letting it run by itself. A beginner can get it working in about half an hour.

Step one, power on and pair. Long-press the power button on the body for two seconds; when the indicator goes from blinking red to solid green, it's up. Install its app on your phone, search via Bluetooth, and you'll see a device name starting with ev¹ — tap it. The first pairing takes a dozen-plus seconds; the screen spinning is normal, don't tap repeatedly.

Step two, don't rush into learning mode — walk it manually once first. This was the first pit I stepped in. I went straight into learning mode, and the route the vehicle recorded came out all crooked, because I myself wasn't used to the vehicle's steering effort. Manual mode is the traditional electric wheelchair setup: push the joystick forward to go, pull back to stop, steer by pushing left and right.

I tried two approaches. First run it manually once or twice to get a feel for the vehicle, then enter learning mode — the recorded route is noticeably cleaner; going straight into learning mode, the first attempt will most likely need re-recording. On the same roughly 400-meter campus route, the former passed after two recordings, while the latter was still correcting deviations after four.

Step three, enter learning mode and walk it once. In the app choose Route → New → Learning mode. The vehicle switches to follow state, you walk about one to two meters in front of it, and it follows your trajectory. If the trajectory line on screen is green, it's tracking; if it turns yellow, it's about to lose the target and you need to walk slower.

Step four, save at the endpoint. Tap End learning, give it a name, like "Cafeteria to Dorm." It will ask whether to return to the start and replay once — choose yes. During replay, stand aside and don't move, watch it walk by itself. This part is the most like acceptance testing — whether the trajectory hugs the path and whether obstacle avoidance works, you can tell at a glance.

Step five, takeover. At any time, pushing the joystick or pressing the stop button takes over. This is a must-have, otherwise you wouldn't dare use it.

Pitfall section, a few specifics.

One, walking too fast. Learning mode relies on the vehicle's 360-degree sensor suite to recognize you; if you walk too fast and go beyond the follow range, it will stop and wait, and the trajectory breaks. Beginners tend to do this — walk two steps and look back.

Two, reflective ground. There's a stretch in the campus paved with polished tiles; when the sun hits it, the lidar treats it as a void and the vehicle suddenly slows down. This is a common issue with this kind of sensor — just go around or hug the edge.

Three, turning radius. If you turn right against the wall corner during learning, what it records is a wall-hugging trajectory, and during replay being too close to the wall easily triggers obstacle avoidance and a stop. Leave about half a meter of margin, and you'll save yourself a lot of trouble later.

Four, slopes. Before a ramp, remember to switch the chassis to the corresponding mode, otherwise going over the bump will jolt.

After getting it working, my biggest takeaway is that the bottleneck of this thing isn't the hardware, it's whether the route itself is clean. Same principle as risk control: no matter how beautifully the rules are written, if the input is dirty, the false positives won't come down. When you record a route, if you casually take shortcuts and hug walls, it all gets paid back to you during replay.

Next I plan to stitch the cafeteria and dorm segments together and run the whole thing, to see how it handles the junction points. As for whether OTA can be applied directly to already-learned routes or whether they need re-recording, that'll have to wait until the update rolls out. I haven't dared try traffic-light intersections yet — for now, let's get this 400-meter campus stretch running smoothly.