How do I install the Reconstructing Collision Worlds skill?

Run npx skillmds@latest add graph-robots/reconstructing-collision-worlds in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Reconstructing Collision Worlds safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Reconstructing Collision Worlds?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Reconstructing Collision Worlds free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under Apache-2.

Who published Reconstructing Collision Worlds?

graph-robots (@graph-robots) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Reconstructing Collision Worlds by graph-robots

Name: Reconstructing Collision Worlds
Author: graph-robots

Reconstructing Collision Worlds

Reconstructs a planner collision world from calibrated RGB-D views alone, excluding the grasp target, the robot, and a perceived fixture mask, dropping depth-edge slivers and invalid-depth sheets, and carving the free space an intentional-contact goal needs as an approach tube along a fixture axis or a corridor through a lid aperture. Use when a held-object motion must be planned against obstacles no simulator or CAD scene supplies, before registering the object and planning the carry.

graph-robots add6d8c 2 files · 18.2 KB Updated Sep 17, 2026

File contents

reconstructing-collision-worlds

Run once after perception and before grasping, with the same observation the perception skills used. build_collision_world merges depth from the selected views (camera_names, comma-separated; empty = all) into alpha-shape meshes through geometry.build_world_config and returns a WorldConfig plus the mesh names.

What is excluded

The grasp target by its perceived target_mask in mask_camera_name (default overhead), and by segmenting target_description in every other selected view so the object does not become a static obstacle where it later rotates in the hand.
The robot, by text segmentation per view and by motion.get_robot_collision_spheres when the connector provides it (a connector without it falls back to the visual masks).
A perceived fixture_mask from the mask camera, unless it is a full-frame placeholder that would erase the world.

What is carved

An approach tube of 55 mm radius along outward_sign * fixture_axis ending 12 mm behind fixture_tip, when both are given. Use outward_sign: -1 for an axis that points into the fixture (an aperture's inward axis) and 1 for one that points toward its free side (a shaft).
A vertical corridor of corridor_radius through corridor_center between corridor_rim_z - 35 mm and +180 mm, when the centre and a positive radius are given. The radius is the caller's measurement (for example a fraction of the perceived aperture radius); the script adds no margin.

Cleaning and packing

Sparse, nearly zero-thickness sheets under 15 cm are dropped; components lying entirely below sheet_floor_z or spanning more than sheet_max_extent_m are invalid-depth returns and are dropped too (0 disables either gate). voxel_size trades surface ripple for detail. pack_meshes: true folds every component into one perceived_scene mesh for planners whose scene cache admits few named obstacles.

Boundaries

No simulator state, CAD model, or fixed world coordinate is read.
The script raises when no mesh survives; the graph routes that to failed.

---name: reconstructing-collision-worldsdescription: Reconstructs a planner collision world from calibrated RGB-D views alone, excluding the grasp target, the robot, and a perceived fixture mask, dropping depth-edge slivers and invalid-depth sheets, and carving the free space an intentional-contact goal needs as an approach tube along a fixture axis or a corridor through a lid aperture. Use when a held-object motion must be planned against obstacles no simulator or CAD scene supplies, before registering the object and planning the carry.license: Apache-2.0---# reconstructing-collision-worldsRun once after perception and before grasping, with the same observationthe perception skills used. `build_collision_world` merges depth from theselected views (`camera_names`, comma-separated; empty = all) intoalpha-shape meshes through `geometry.build_world_config` and returns a`WorldConfig` plus the mesh names.## What is excluded- The grasp target by its perceived `target_mask` in `mask_camera_name` (default `overhead`), and by segmenting `target_description` in every other selected view so the object does not become a static obstacle where it later rotates in the hand.- The robot, by text segmentation per view and by `motion.get_robot_collision_spheres` when the connector provides it (a connector without it falls back to the visual masks).- A perceived `fixture_mask` from the mask camera, unless it is a full-frame placeholder that would erase the world.## What is carved- An approach tube of 55 mm radius along `outward_sign * fixture_axis` ending 12 mm behind `fixture_tip`, when both are given. Use `outward_sign: -1` for an axis that points into the fixture (an aperture's inward axis) and `1` for one that points toward its free side (a shaft).- A vertical corridor of `corridor_radius` through `corridor_center` between `corridor_rim_z - 35 mm` and `+180 mm`, when the centre and a positive radius are given. The radius is the caller's measurement (for example a fraction of the perceived aperture radius); the script adds no margin.## Cleaning and packingSparse, nearly zero-thickness sheets under 15 cm are dropped; componentslying entirely below `sheet_floor_z` or spanning more than`sheet_max_extent_m` are invalid-depth returns and are dropped too (0disables either gate). `voxel_size` trades surface ripple for detail.`pack_meshes: true` folds every component into one `perceived_scene` meshfor planners whose scene cache admits few named obstacles.## Boundaries- No simulator state, CAD model, or fixed world coordinate is read.- The script raises when no mesh survives; the graph routes that to `failed`.

Frequently asked questions