earakely-scale commited on
Commit
0fca2e2
·
verified ·
1 Parent(s): 798e221

MuJoCo grades, Genesis gives a second opinion; a knife-block set and its MJCF + OBJ export

Browse files
README.md CHANGED
@@ -101,6 +101,31 @@ The scale tasks are the open problem. Even with `source_photo`, neither model re
101
  That matches the VLM-to-CAD step itself, which misjudged these same objects' size from these same photos.
102
  Measuring metric size from photos needs a metric source: depth, a known-size reference, or a person.
103
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
104
  ## With a contributor in the loop
105
 
106
  The scale tasks fail because the real size isn't in the asset. The `contributor-*` tasks put a contributor in the loop.
@@ -119,10 +144,10 @@ agent's edits, and shows the object as it was at each moment.
119
 
120
  *`contributor-cad-mug`, recorded in Chrome: the contributor lifts the mug (the red cylinder is a 330 ml can, for scale),
121
  says it's tiny, gives its real size when the agent asks, checks the fix on the tilting table, and approves it. Stretches where the agent is
122
- working play at 8× speed. The other recordings: [the milk frother's loose ring](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/videos/cad_milk_frother.mp4)
123
- and [the cupboard's jammed doors](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/videos/rcad_cupboard.mp4).*
124
 
125
- A script plays the contributor in these 14 sessions, not a person. It knows each object's real size and weight from the
126
  Google scan the CAD was built from, judges each fix only from what the page shows it, and never sees the hidden checks.
127
  Some sessions went through the page in Chrome with `scripts/record_session.py`, and the rest through the page's HTTP
128
  API with `scripts/contributor.py`. The six newest sessions also carry each action's Genesis trace (`traces`): every visual
@@ -137,8 +162,10 @@ geom's pose per frame. The agent is Claude Code with `claude-sonnet-5-5`.
137
  | `contributor-cad-frother` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `remove_part(part=cord, restore=False)` | `cad_milk_frother` v1 |
138
  | `contributor-cad-frother` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `remove_part(part=cord, restore=False)` | `cad_milk_frother` v2 |
139
  | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | scripted | `set_scale(factor=1.4)`; `set_shell(part=body, thickness=0.004)`; `set_shell(part=body)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v2 |
 
140
  | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | browser, recorded | `set_scale(factor=1.38)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v3 |
141
  | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | browser, recorded | `set_scale(factor=1.4)`; `set_shell(part=body, thickness=0.004)`; `set_shell(part=body)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v4 |
 
142
  | `contributor-rcad-cupboard` | 1.00 / 1.00 | 1.00, approved | scripted | `set_self_collision(enabled=False)` | `rcad_cupboard` v1 |
143
  | `contributor-rcad-cupboard` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `set_self_collision(enabled=False)` | `rcad_cupboard` v2 |
144
  | `contributor-rcad-cupboard` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `set_self_collision(enabled=False)` | `rcad_cupboard` v3 |
 
101
  That matches the VLM-to-CAD step itself, which misjudged these same objects' size from these same photos.
102
  Measuring metric size from photos needs a metric source: depth, a known-size reference, or a person.
103
 
104
+ ## MuJoCo grades, and approved objects ship as MJCF and OBJ
105
+
106
+ The objects ship as MJCF and OBJ for MuJoCo, so the workbench grades in MuJoCo. Every contributor action, every agent
107
+ probe by default, and every hidden check runs there. Genesis runs the same checks as a second opinion, and each graded
108
+ episode lists where the two engines disagree. Approving an object exports it as it ships:
109
+
110
+ - `scene.xml`, which attaches each object from its own MJCF file;
111
+ - one MJCF per object, with an explicit `<inertial>` per body and friction per geom;
112
+ - the OBJ meshes, with collision as convex pieces;
113
+ - `report.json`.
114
+
115
+ The export is loaded back and stepped in MuJoCo before it counts. Episodes and sessions recorded before this change
116
+ were graded in Genesis.
117
+
118
+ ![The knife block's validation: knives dropped into their slots, the checks in MuJoCo and Genesis, the tipping margin](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/knife-block-report.png)
119
+
120
+ **A set: a knife block with six knives.** A beech block on rubber feet with six free knives that sit in its slots. Its
121
+ probes ask what a robot would: do the knives stay seated, drop in, and pull free without lifting the block, and how far
122
+ can the table tilt? As delivered in `knife-block-hull` and `contributor-knife-block`, the chef's and bread knives were each
123
+ given to MuJoCo as one collision mesh. MuJoCo collides a mesh as its convex hull, so those knives are 21 mm wide and
124
+ can't enter their 4.5 mm slots, and the set fails four of five checks. With a convex hull per part (`set_collision`) it
125
+ passes all five, and Genesis agrees on every check it runs. Its static tipping margin is 13.4° to the back, where the
126
+ block and handles lean, and 22–37° in other directions. The [Space](https://huggingface.co/spaces/earakely-scale/GenesisWorkbenchEnv-AgentEnv)
127
+ replays every probe in 3D. Claude Code on `claude-sonnet-5-5` fixed it alone in 37 s, because `inspect` shows each knife's collision thickness beside its slot's width. [`exports/knife_block.zip`](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/exports/knife_block.zip) is the fixed set as MJCF and OBJ.
128
+
129
  ## With a contributor in the loop
130
 
131
  The scale tasks fail because the real size isn't in the asset. The `contributor-*` tasks put a contributor in the loop.
 
144
 
145
  *`contributor-cad-mug`, recorded in Chrome: the contributor lifts the mug (the red cylinder is a 330 ml can, for scale),
146
  says it's tiny, gives its real size when the agent asks, checks the fix on the tilting table, and approves it. Stretches where the agent is
147
+ working play at 8× speed. The other recordings: [the knife block's two knives that won't go in](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/videos/knife_block.mp4),
148
+ [the milk frother's loose ring](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/videos/cad_milk_frother.mp4) and [the cupboard's jammed doors](https://huggingface.co/datasets/earakely-scale/GenesisWorkbenchEnv-AgentEnv/resolve/main/videos/rcad_cupboard.mp4).*
149
 
150
+ A script plays the contributor in these 16 sessions, not a person. It knows each object's real size and weight from the
151
  Google scan the CAD was built from, judges each fix only from what the page shows it, and never sees the hidden checks.
152
  Some sessions went through the page in Chrome with `scripts/record_session.py`, and the rest through the page's HTTP
153
  API with `scripts/contributor.py`. The six newest sessions also carry each action's Genesis trace (`traces`): every visual
 
162
  | `contributor-cad-frother` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `remove_part(part=cord, restore=False)` | `cad_milk_frother` v1 |
163
  | `contributor-cad-frother` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `remove_part(part=cord, restore=False)` | `cad_milk_frother` v2 |
164
  | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | scripted | `set_scale(factor=1.4)`; `set_shell(part=body, thickness=0.004)`; `set_shell(part=body)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v2 |
165
+ | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | scripted | `set_scale(factor=1.4)`; `set_shell(part=body, thickness=0.005)`; `set_shell(part=body)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v5 |
166
  | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | browser, recorded | `set_scale(factor=1.38)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v3 |
167
  | `contributor-cad-mug` | 0.50 / 0.75 | 1.00, approved | browser, recorded | `set_scale(factor=1.4)`; `set_shell(part=body, thickness=0.004)`; `set_shell(part=body)`; `set_inertial(link=body, mass=0.3)`; `set_scale(factor=1.516)`; `set_inertial(link=body, mass=0.3)` | `cad_mug` v4 |
168
+ | `contributor-knife-block` | 1.00 / not run | 1.00, approved | browser, recorded | `set_collision(body=chef_20cm, mode=parts)`; `set_collision(body=bread_20cm, mode=parts)` | `knife_block` v1 |
169
  | `contributor-rcad-cupboard` | 1.00 / 1.00 | 1.00, approved | scripted | `set_self_collision(enabled=False)` | `rcad_cupboard` v1 |
170
  | `contributor-rcad-cupboard` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `set_self_collision(enabled=False)` | `rcad_cupboard` v2 |
171
  | `contributor-rcad-cupboard` | 1.00 / 1.00 | 1.00, approved | browser, recorded | `set_self_collision(enabled=False)` | `rcad_cupboard` v3 |
exports/knife_block.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c5ef0d69c7de07e9a8ae7dbabf770590e4f12e7af14d85f7a9bb332281588aab
3
+ size 30763
knife-block-report.png ADDED

Git LFS Details

  • SHA256: 0e826790dbf796072f1834f31d7267912eb5d764d48ef4283bb0a93cf27984eb
  • Pointer size: 131 Bytes
  • Size of remote file: 326 kB
sessions.jsonl CHANGED
The diff for this file is too large to render. See raw diff
 
videos/knife_block.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:54794369c926d33c9f5bfd69cf46f1e9f128f279fd65987f402c89755be1e1f6
3
+ size 1014475