Turning a Wheeled Chassis Into Blocks: Fun to Tear Down, But You Route the Wiring Yourself

This extendable modular mobile platform from Lingsi — its structure and interfaces are made for people willing to get hands-on, and I'm satisfied with the disassembly experience; but the accompanying software and docs are still at the engineering-prototype stage, so teams expecting out-of-the-box will suffer.

Last month I spent two days at a client's site. Their production line wanted to add mobile grasping; the arm and hand are ours, and the base was originally a certain all-in-one from another vendor — once the arm was mounted the whole machine wobbled, and changing the structure meant shipping the whole thing back to the factory. This time we tried Lingsi's modular base. I disassembled and reassembled it together with their engineers.

At first glance it's a white wheeled base, parked on a wooden floor, four wheels, battery and drivers in the middle, three layers of aluminum-profile extension layers stacked on top, a standard hole pattern on the top surface, and power and communication interfaces visible on the side. So-called modular means treating the base as a powered building-block platform — stack whatever you want on top yourself. This also connects to my earlier judgment that mobility plus manipulation will eventually have to merge. In that Gege post a while back I said it sidestepped grasping with a dual-cabin design, but how base and manipulation really merge still depends on this kind of reconfigurable foundation.

The disassembly was simpler than I thought. A few hex screws fix the extension layers together, two people can lift one off, and the holes have locating pins so you don't have to re-align when putting it back. This matters a lot to me — on my last project, every arm swap meant recalibrating hand-eye, which took half a day each time. Here, swapping the upper layer leaves the base's odometry and inertial measurement unit untouched. Odometry is the base estimating how far it's traveled, and the IMU measures attitude — once those two move, all the upper-layer calibration is wasted. I understand this is the real meaning of "extendable": the point is you don't have to start over when you add things.

Wiring and power-up weren't as smooth. There's no reserved cable channel between the extension layers, so power and bus cables have to be threaded through the profile gaps yourself. I temporarily tied them with zip ties — honestly not pretty; for long-term use you should make your own harness. After power-up, connect to the host computer — the interface is typical engineering software, node tree on the left, a pile of parameters on the right, all named with abbreviations, a bit baffling at first glance. Fortunately the on-site engineer was familiar and enabled the base with a couple of clicks.

The controller's joystick feel is so-so, response keeps up. I had it walk a stretch in a straight line and come back, watching the odometry curve — on the way back it drifted a bit, not outrageous, good enough on flat ground. The biggest advantage of wheels is you don't worry about it standing steady — none of that quadruped balancing business to deal with. These past few days I happen to be testing the Deep Robotics Lite3, and the contrast is obvious: the difficulty with legs is stability, while with wheels the difficulty is all in the upper layer.

The pleasant surprise was rigidity. After mounting the arm I pushed on it, and the wobble was much less than that all-in-one from before — probably related to the profile thickness and fixing method. This is crucial for grasping: if the base is a bit soft, the hand's positioning accuracy is all wasted, and no matter how high the tactile array resolution, it's useless.

Next I plan to mount the tactile acquisition box too, run it on an independent communication line, and also test long-distance straight-line drift. Haven't tried outdoors yet — wheels need to go over thresholds at an angle, and unstructured ground probably isn't its arena.

Who it suits: teams doing mobile manipulation research that need to repeatedly change the upper structure, or doing prototype validation on a production line where the base stays put but the top changes often. Who it doesn't suit: those who want out-of-the-box, complete docs, and a friendly interface; and if your site has steps or uneven ground, don't count on it.

Whether a base is good or not, in the end it comes down to whether you still dare to move it after you've stacked things on.