What changes when the X2 has to find the part itself

A fixed-base arm never has to ask where the part is. It is bolted to a position, the part arrives on a line at a known height, and everything the specification sheet promises is measured from that fixed point. A mobile platform gives that up on purpose, and the moment it does, a different set of numbers starts to matter.

The X2 Ultra is the useful comparison because AGIBOT lists it in the same catalogue as the stripped-down X2 Basic, and what separates them is the sensing and compute that navigation needs. The Ultra carries an NVIDIA Orin NX board rated at 157 TOPS, moves at up to 2 m/s, and runs for 2 hours between charges, with 3 kg per arm to spend once it arrives. The Basic does without the LiDAR and depth camera the Ultra adds, so it is the Ultra that has to solve a problem the Basic is not equipped for: find its own way to the part before it can do anything with it.

Compute is spent before the arm moves.

157 TOPS on the Orin NX board is not idle headroom for a bigger neural net later; a mobile platform spends a meaningful share of it on localisation and obstacle avoidance before a single grasp is planned. A fixed-base arm's controller spends none of its budget finding out where it is. That is the honest cost of removing the bolts, and it is worth asking a vendor how much of a published TOPS figure is already committed before task planning starts.

2 m/s is a travel speed, not a cycle speed.

It tells you how fast the platform crosses a floor between stations, not how fast it completes a pick. A cell designed around a fixed arm can be laid out so cycle time is dominated by the grasp. A cell designed around a mobile platform has travel time added on top, and that travel time is exposed to everything on the route: a pallet in the aisle, a door that opens the wrong way, a person crossing at the wrong moment. The number is real. What it measures is narrower than it sounds.

2 hours of runtime is the number that sets the shift plan.

A fixed-base arm runs until someone switches it off. A mobile platform has to come back, and 2 hours between charges means the swap or charging routine is part of the deployment from day one, not an afterthought once the pilot works. That is a layout question as much as a power question: somewhere for the platform to return to, and a plan for what covers the floor while it does.

3 kg per arm is the number that decides whether autonomy was worth having.

A robot that can find its own way to a part but cannot lift the part once it gets there has solved the wrong half of the problem. The Ultra's payload figure is what a buyer should check first, before speed or compute, because navigation only pays for itself if the arm on the other end of the journey can actually do the job waiting for it.

What the catalogue does not say.

The catalogue gives no positioning accuracy figure for the Ultra's navigation, nor a maximum traversable gap or step height. Those are exactly the numbers that decide whether a given floor plan works without a site visit, and their absence is worth naming rather than assuming: ask for them before specifying a cell around this platform, not after.

Autonomy is not free. It is traded for the compute budget, the travel time, and the return-to-base routine listed above, and the trade is only worth making if the payload at the end of it can carry the part. That is the order to check the numbers in, not the order they appear on the sheet.

We can talk through what a mobile cell built around these figures would need before anything is specified.