MARSAD / CONSOLIDATED CATALOG / #016

Child breaks free during a roadside reprimand

Awaiting your 3D scene approval

Counterfactual proposal: a parent reprimands a child while holding one hand on the south sidewalk. The upset child breaks free and runs across the road. The parent follows on foot, reunites with the child on the north sidewalk, and accompanies the child away.

Original description: 前方有家长在路边训斥孩子，孩子情绪激动突然跑开，挣脱大人的手冲向马路。

47 sFull continuous rehearsal

2Critical actors and props

941Exported time steps checked

0New Image 2.5 outputs

Speed Follow actors

Loading the full animated scene…

Critical moments

What to verify

Verify exactly one parent and one child persist throughout all 941 playback samples, including -7 s and 40 s.
Verify the initial hand contact is reachable without torso overlap, ends at -1 s, and does not pull either actor discontinuously.
Verify the child's running phase remains within 2–4 m/s apart from acceleration and braking; the parent's walking phase remains within 0.6–1.7 m/s apart from start and stop transitions.
Verify all waiting and final footprints lie inside the wheelchair map's sidewalk polygons; only the authored crossing occupies the road.
Verify ordinary walking separation is at least 0.6 m, excluding the explicitly staged initial handhold.
Verify ego remains stopped at x=12 until 13 s and neither person can re-enter its buffered corridor during the resumed passage.
Require human approval of the full timeline, source-background conflicts, and the proposed 1.5 s review view before any image generation.

Review notes

Reviewed the child leaving the parent, crossing ahead and reuniting on the opposite sidewalk. Adult and child remain distinct and the parent follows later. Hand-holding at the initial state is an approximate relationship.

Check action, relative position, facing, posture and visibility. Approximate offsets are acceptable. Sampled visual checks do not prove every contact or continuous collision clearance.

Approval stops here. Please identify this scene as #016 when approving it or requesting revisions. Image 2.5 generation and training remain paused for this new catalog phase.

Rendered inspection views

Event, approach and final views rendered from the exact committed Higgsfield Blender source. These are procedural 3D previews, not generated training photographs.

-7 s · Parent holds the child's hand and gestures sternly on the south sidewalk · REVIEW_WORK_AREA

-1 s · Child releases the handhold and accelerates toward the road · REVIEW_WORK_AREA

0 s · Child reaches the curb and steps into the roadway · REVIEW_WORK_AREA

1.5 s · Child runs through the stopped ego's forward corridor · REVIEW_WORK_AREA

4.5 s · Child stops on the north sidewalk · REVIEW_WORK_AREA

10.5 s · Following parent reaches the north sidewalk · REVIEW_WORK_AREA

12 s · Parent and child begin walking east together, entirely on the sidewalk · REVIEW_WORK_AREA

13 s · Ego may resume after the corridor remains clear · REVIEW_WORK_AREA

40 s · Both remain together at their final sidewalk positions · REVIEW_WORK_AREA

1.5 s · Original calibrated forward camera at the critical event · CAM_F0

Original NAVSIM camera context

Original NAVSIM CAM_F0 — Original NAVSIM anchor · CAM_F0. This is the untouched camera context, not the synthetic scene.

Original NAVSIM CAM_R0 — Original NAVSIM anchor · CAM_R0. This is the untouched camera context, not the synthetic scene.

1 conflicting background vehicles were moved to clear holding positions for this counterfactual rehearsal. Original photos remain unchanged. Before later inpainting, any overlap with original-photo traffic must be resolved. Background positions in this 3D rehearsal are not an exact replay of the photographs.

After scene approval

The working pipeline will edit each original NAVSIM camera photograph with Image 2.5. Its same-time Higgsfield view will provide position and posture, annotated with facing direction and current speed. Temporal image requests will cover the full event with a 1 Hz base plus critical moments.

Assumptions and source provenance

Image #1 supplies coarse street architecture, sidewalks, trees, and parked-car context. The parent, child, gestures, and trajectories are counterfactual additions, not observed events.
Playback spans -7 through 40 s inclusive at 20 Hz: 941 samples, finite and non-looping. Use bounded, ground-following C1 interpolation, tangent-aligned turns, smooth starts and stops, and exact stationary holds; do not introduce spline overshoot.
Coordinates use the supplied wheelchair polygons. Around this crossing the south sidewalk is near y=-6.2 and the north sidewalk spans approximately y=5.3 to 8.1. All human roots use z=0.
Ego is controlled separately and is not duplicated in the actor list. Assume a 4.5 m by 1.9 m ego, starting at x=4, y=0 at -7 s with speed 4 m/s, braking smoothly to x=12 by -3 s. After release it may accelerate smoothly to 2 m/s along y=0.
The buffered ego corridor is y=-1.5 to 1.5. Reuniting on the sidewalk, rather than a sudden roadside reversal, supplies the proposed resolution.
Precise background parked-car footprints and curb heights are not supplied and require human conflict review. Approval is pending; later Image 2.5 must edit original Image #1, using this rehearsal only for actor position and posture.

Fresh GPT-6 motion plans use semantic descriptions from the pinned consolidated catalog and NAVSIM environments. No upstream nuPlan vectors, trajectories or images are used. The four source environments are reused; 82 scenario descriptions do not mean 82 independent locations.

Higgsfield revision 1. The portable GLB combines export clips without changing geometry or animation samples.

This review uses Higgsfield revision 1 and the exact exported scene snapshot.

Higgsfield workspace (latest revision) · Download animated GLB · GPT-6 scene plan · Visual review evidence · Motion checks · Approval status