Skip to content

armgit / moss

Public Imported from GitHub
0

MOSS: a small, mostly 3D-printed litter-picking rover. Cleaner streets, one maker at a time.

Python

build log: measured mass, model mass and camera position, stated as what they are

metrox67a331e41e4b
diff --git a/docs/build_log.md b/docs/build_log.md
index 1302869..fb0d680 100644
--- a/docs/build_log.md
+++ b/docs/build_log.md
@@ -126,5 +126,7 @@ The CPU MuJoCo simulator behind the Jev demo is now published in the demo reposi
Same day, a [V0.4 appearance overlay](https://github.com/metrox-eth/moss-jev/tree/main/live/model/visuals_v04) followed: the saved V0.4 hull, side assemblies and camera envelope, applied by a small script that keeps the host's masses, joints, cameras and contacts untouched. Two things to know if you import it. The CAD belts sit 266 mm apart while the legacy collision boxes are 244 mm apart; the overlay does not change the drive model, so a physically consistent V0.4 driving variant is still to do. And the visual meshes are decimated STL (5,000 to 6,000 triangles per part, no smoothed normals), which shows as facets in a renderer; a finer export is planned.
+Numbers worth writing down while they are fresh, because a model copied often enough starts to look like a measurement. The physical prototype was weighed at 3.5 kg without the arm, cover (140 g) and bin (180 g) included, so about 3.18 kg bare. The 6.14 kg in the first MJCF export is an inertia computed from solid geometry, not a weighing. The camera's nominal front face sits at (156, 0, 75) mm in the rover frame, level; the 95 mm sometimes quoted is a mounting spacing, not its height. The arm and gripper are about 680 g by model, plus 50 g allowance, not weighed; the vertical centre of gravity, 75 mm, is estimated, and the "CG above the middle roller" observation was made without cover, bin or arm, so it does not apply to a loaded bin.
+
Within the day, another lab imported the body into its own simulator and started training a can-scooping policy by reinforcement learning on it, 32 environments in parallel. The base in our model is a simplified slide, not tracked-ground dynamics; that limitation is theirs to work around too.