FAQ for engineers

No — and this is fundamental. The drawing is a coordinate system for the report, not a map for the legs. The robot builds its own real-time 3D map (SLAM: onboard LiDAR + inertial unit + cameras) and localizes within it.

Obstacles, people, materials, a wall that didn't exist yesterday — all seen by live sensors, not taken from a design that on site always lags reality.

The first visit is a teach-run: an operator walks the robot along the route, stairs included. The robot memorizes the 3D trajectory with checkpoints — and then repeats it autonomously, pass after pass.

That is exactly why the tenth report is comparable to the first: the route is physically the same. Checkpoints are conveniently assigned right on the floor plan — after binding, the robot matches them to its map.

Not from the plan — a 2D drawing has no heights. Depth cameras and LiDAR see steps as geometry, and the quadruped's locomotion controller is trained to place feet on steps (rated: steps up to ~20 cm, slopes up to 40°).

A multi-floor map is a graph of floors: "floor 3 → stair segment → floor 4", and the links are recorded during the teach-run. At the monolithic stage, without finished flights, floors are captured separately with manual transfer of the robot.

Yes. The comparison happens not "in the dog" but in the WOLKER cloud:

• The robot captures a point cloud of the floor (~1.2 cm accuracy), knowing nothing about the design.
• The cloud is aligned to the drawing's coordinate system — by axes, checkpoints, or best-fit to walls.
• In-plane deviations: a horizontal slice of the cloud at a set height → "as-built floor plan" → overlaid on DWG polylines: wall positions, room dimensions, opening sizes and positions — in numbers.
• Wall verticality: planes are fitted into the cloud, tilt measured from true vertical (gravity known from IMU) — in mm over floor height, against SP 70.13330 tolerances. No drawing needed for this.
• Openings: window and door edges extracted from the cloud → actual sizes and levels vs. design.

The result can come out as a PDF overlay on top of the project sheet: as-built in red on the design drawing. With BIM/IFC the comparison is fully volumetric; with only 2D AutoCAD — floor-by-floor comparison plus independent verticality, which is enough for technical supervision.

Input: DWG/DXF, IFC (BIM), PDF underlays.

Output: E57/LAS (point cloud), DXF/PDF overlays, deviation schedules in CSV/XLSX — compatible with any CAD/BIM environment.

Repeatability: the same route and sensors on every pass, so deviation is visible in dynamics — what changed since yesterday, what after the concrete pour, what after formwork removal.

The support-surface map is built continuously. A hole is the absence of beam return where a floor should be. Elevator shaft, ventilation duct, slab edge, floor opening — identical prohibition. The rule is hard: no floor data = no-go there.

All shafts and openings are known from DWG/IFC in advance — a keep-out layer is built automatically: a 1.5–2 m buffer from the slab edge, zones around each shaft. The robot plans no trajectory there, even if sensors "see floor" — film or plywood over an opening is visible, but its load capacity is visible to no one.

The robot follows the recorded trajectory with ~0.5 m tolerance; everything outside the corridor is prohibited by definition. Routes are laid through the interior zones of the floor, not along the edge.

Loss of localization, dust, darkness, map-vs-reality divergence → stop and alert the operator, not "I'll try to get through". Abnormal tilt or leg subsidence → protective stop.