Debug Tool: Heightmap Replacement

Overview

Checks rayrai depth images for stale geometry after removing heightmaps and primitives. The scene places a PD-controlled ANYmal on a flat terrain patch and renders its front and rear depth cameras into separate ImGui panels, and draws the pixels of each camera projected into the world as small instanced cubes. Press Space to create a visibly different heightmap with broad mounds that shift across both camera frames, then replace randomized populations of simple primitives and delete the old scene.

Target

CMake target: rayrai_heightmap_replacement. Source: examples/tools/debug/rayrai_heightmap_replacement.cpp.

Run

Build with RAISIM_DEBUG_TOOLS=ON as described in Regression and Debug Tools, then run the build-tree executable:

./build-debug/examples/rayrai_heightmap_replacement

On Windows, run build-debug\bin\rayrai_heightmap_replacement.exe instead. This tool renders in process with rayrai and does not need rayrai_tcp_viewer.

How to reproduce

  1. Watch the mounds in the front and rear depth images shown at the upper left.

  2. Tap Space several times. Each press removes the current heightmap and creates a new generation with a random number of boxes, spheres, and cylinders in each camera frame.

  3. Compare the depth image with the current colored terrain in the main view. A mound or primitive from an earlier generation indicates stale deleted geometry in the depth pass.

  4. The projected pixel cubes must lie on the current terrain and primitives. Cubes floating in the air or buried in the ground show that the depth image does not match the scene.

Details

  • Loads the sensored ANYmal used by sensor_suite and selects its front and rear depth cameras.

  • Uses PD gains and a nominal standing target to hold the twelve leg joints.

  • Keeps a radius of 1.15 m around the robot exactly flat, with a smooth transition to rolling terrain.

  • Moves broad near-field mounds left and right across both camera views on every terrain generation.

  • Places two to nine deterministic pseudo-random static boxes, spheres, and cylinders in each camera’s near field.

  • Projects each depth image into world-frame points with DepthCamera::depthToPointCloud and draws every sixth pixel in each direction as a small cube, one InstancedVisuals batch per camera in the color of its frustum (orange front, blue rear), darker with distance. The batches are not detectable, so the cameras never see the cubes.

  • Allocates the new heightmap and primitives before calling World::removeObject on the old scene, guaranteeing distinct object addresses while still exercising deletion.