Datasets:
HelenPeng
/
ExeWorldBench

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
596
**10** = The mechanical structure is complete and fully correct, with all major components fully visible in the frame. Geometry, topology, connectivity, spatial relationships, and relationships between mechanical components closely match the reference. All contacts, joints, engagements, and relative placements are mech...
**9** = The mechanical structure and relationships between components are fully correct and clearly visible, with no physically implausible relationships. All major components are fully visible. Only very minor shortcomings exist in aesthetics, lighting, materials, 3D presentation, or camera placement, while the overal...
**8** = The mechanical structure and relationships between components are correct and physically plausible, with no clear interpenetration, unintended overlap, floating components, incorrect engagement, or other mechanically implausible relationships. All important relationships can be visually verified. The main limit...
**7** = No clear physically or mechanically incorrect relationship is visible, but some relationships between components are difficult to verify because of camera angle, occlusion, or visual-quality limitations. Most important connections and spatial relationships remain observable, with only limited uncertainty.
**6** = The visible structure remains mechanically plausible, with no clearly observed interpenetration, floating components, incorrect engagement, or other physical errors. However, several important relationships between components cannot be reliably verified because of camera angle, occlusion, poor visibility, or re...
**5** = The overall structure remains mostly correct, but a small number of clear physical or mechanical relationship errors appear, typically limited to 1–2 local cases or a few frames. Examples include two components visibly intersecting, a ball or component unnaturally floating without support, components that shoul...
**4** = Physical or mechanical relationship errors become more frequent or substantial, with multiple cases of interpenetration, floating components, incorrect engagement, implausible connections, or other invalid component relationships. Important components may also be substantially cropped or occluded, although the ...
**3** = Severe structural or physical errors are widespread across multiple components. Interpenetration, floating, incorrect connections, implausible component relationships, or incomplete components occur extensively, or the component modeling, mechanical topology, and connectivity themselves are clearly incorrect, p...
**2** = Most of the mechanical structure contains severe problems. Many major components are missing, severely malformed, misplaced, disconnected, or unobservable, with only limited structure corresponding to the reference.
**1** = Only minimal recognizable structure related to the reference remains. Most major components and mechanical relationships are missing or severely incorrect, and the result largely fails to form a meaningful mechanical device.
**0** = No meaningful valid mechanical structure corresponding to the reference is present or observable.
==================================================
2. EXECUTABLE CAUSAL MECHANISM
==================================================
Evaluate whether the generated world forms a real, continuous, complete, and mechanically plausible dynamic execution process.
This dimension contains two aspects:
1. **Mechanism Execution:** whether individual mechanical objects and mechanisms execute correctly;
2. **Causal Connectivity:** whether different mechanical stages form a continuous execution chain through real, effective, and mechanically plausible interactions.
Both aspects jointly determine the final 0–10 score.
Correct local motions alone are insufficient if they do not form a real mechanical chain. Likewise, a correct event sequence is insufficient if local mechanisms move incorrectly or violate physical principles.
### A. Mechanism Execution
For each key mechanism participating in execution, check:
* Whether its motion type, direction, range, and magnitude are reasonable;
* Whether it respects its mechanical degrees of freedom, pivots, axes, hinges, and other constraints;
* Whether its response to physical input has a reasonable direction and relative magnitude;
* Whether there is unexplained acceleration, sudden direction change, or implausible energy gain;
* Whether connected parts remain connected;
* Whether teleportation, jumping, drifting, detachment, penetration, or abnormal deformation occurs;
* Whether motion is temporally continuous;
* Whether the mechanism fully performs its mechanical function and reaches a reasonable resulting state;
* Whether it genuinely participates in the overall execution rather than merely existing or exhibiting superficial motion.
For example, balls should roll, fall, and collide plausibly; levers should rotate around pivots; pendulums should swing around suspension points; hinged structures should remain properly connected.
Exact physical calculations are not required, but motion direction, relative magnitude, and outcome must have a reasonable physical explanation.
### B. Causal Connectivity
For each observed consecutive stage A → B, the sequence should show:
A produces a mechanical output
→ the output physically reaches and acts on B
→ B consequently produces a reasonable mechanical response.
Evaluate the following:
**1. Real Mechanical Triggering**
A → B must occur through an observable physical or mechanical interaction, such as collision, contact, pushing, gravity, constraints, release, catching, or mechanical transmission.
The following do **not** count as valid mechanical triggering:
* B automatically moves at a preset time or animation frame;
* B activates when A enters a region without observable mechanical interaction;
* a “switch event” directly activates a spatially disconnected object without mechanical transmission;
* a predefined logical event directly changes a downstream object's position, velocity, rotation, or state;
* stages simply play sequentially according to a predetermined timeline.
Do not infer triggering merely because “A moves, then B moves.”
**2. Causal Contribution**
A's mechanical action must make a meaningful contribution to B.
As an auxiliary check, ask:
**If A's mechanical action on B were removed, would B still occur in essentially the same way?**
If clearly yes, A may have been bypassed and Causal Connectivity should be reduced.
For example, a ball would enter the next stage even without interacting with a seesaw; a pendulum moves but the downstream object would move anyway; or a component moves without contributing to subsequent execution.
No actual intervention experiment is required. This is only an auxiliary reasoning criterion for identifying bypassed or superficial causal interactions.
**3. Complete and Physically Plausible Mechanical Transmission**
Check whether the contact location, interaction direction, motion strength, and energy transfer are sufficient to explain B's response.
Do not inspect only the instant when triggering begins. Determine whether the observed mechanical interaction is sufficient to explain the **complete downstream motion**.
If the downstream motion physically requires sustained contact, pushing, constraint, or mechanical transmission, that interaction must persist for a sufficient duration. A brief initial contact alone is not enough to explain later motion that would require continued mechanical input.