Repository navigation
feat(assets): a snapshot import places the robot at any heading in the room - #24
Merged
Merged
Conversation
…e room The only open floor beside a reconstructed living room's coffee table faces the robot toward room -Y; the importer assumed +X. Positions and bounds now go through one room-to-robot transform, and kept objects turn with it. Co-authored-by: Cursor <cursoragent@cursor.com>
ZJLi2013
marked this pull request as ready for review
October 22, 2026 14:16
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #23.
What this does
aha3d_snapshottakes--yaw(the robot's heading in the room frame, degrees from +X). Root positions and boundsgo through one room-to-robot transform (
to_robot/box_to_robot; boxes take the bounds of their rotatedcorners) before the existing keep/drop rules, and each kept object gets
yaw_deg = -yaw. The default--yaw 0gives the same output as before. A test checks that turning a scene by 90 degrees and the robot with it gives the
same mapping.
Result: a living room reconstructed from video
The input was an editable room that a coding agent (Codex + GPT-6 Astra) built in Blender from an 8.8 s navigation
clip of the BEHAVIOR Challenge 2025 demos (task
turning_on_radio). The room is in the reconstruction's predictedmetres. The robot stands at the coffee table's +Y end, the only open floor within reach of the radio, so it faces
room -Y (
--yaw -90).table (top 0.389 m becomes 0.775 m). The sofa was dropped by the existing rule because it overlaps the robot's
table by 2 cm. Wall fixtures, the pendant and the structure were dropped as well.
2.2 cm grip depth and 3.35 cm of clearance under the crossbar.
moves 1 mm in 3 s.
Not covered: no grasp was executed. Generated assets stay out of git, so
scenarios/behavior_living_radio.yamlisnot committed.
AI disclosure
Written with Cursor (Claude) acting as the coding agent: it wrote the code, the test, the import and review scripts
and this description, and ran the simulator checks. The room was modelled by Codex with GPT-6 Astra, and the object
and scene reviews were done by GPT-6 Astra. A human approved the design; the code needs human review before merging.