phulocnguyen/PhysQuant-assets

PhysQuant assets

Simulation assets for the PhysQuant (Spatial Understanding) benchmark: the scene backgrounds, tabletop textures, and basket used to collect its datasets and to evaluate policies in closed loop with Isaac Lab (Isaac Sim 6.2.0).

Contents

Path	Contents	Size
assets/scenes/derived/	76 background layers: each places the workcell on open floor, removes ceilings and conflicting furniture, and strips physics	20 MB
assets/scenes/sources/robocasa/	70 RoboCasa kitchens referenced by the layers	9.7 GB
assets/scenes/sources/interior_agent/	6 InteriorAgent apartments (living rooms) referenced by the layers	3.2 GB
assets/textures/libero/	6 LIBERO tabletop textures used for table randomization	15 MB
assets/baskets/, assets/objects_new/	the basket used by the planning tasks	<1 MB
configs/backgrounds.yaml, configs/appearance.yaml	the background catalog and appearance profile matching these files	<1 MB
MANIFEST.json	path, size, and SHA-256 of every file

Background layers reference their source scenes by relative path, so the folder works wherever it is placed. Every reference resolves inside this repository.

Install

Download into the Spatial-Understanding/ folder of a benchmark checkout; the paths match the project layout:

hf download phulocnguyen/PhysQuant-assets --repo-type dataset --local-dir Spatial-Understanding

To fetch only part, use --include, e.g. --include "assets/textures/*" "assets/scenes/derived/*".

Sources and licenses

Part	Origin	License
Kitchen scenes	RoboCasa, converted to USD by the ExtrinsicDexterity project	see RoboCasa's terms
Apartment scenes	InteriorAgent (Kujiale) scenes, converted to USD by the ExtrinsicDexterity project	see InteriorAgent's terms
Tabletop textures	LIBERO (libero/libero/assets)	MIT, Copyright (c) 2023 Lifelong Robot Learning
Derived layers, basket, configs	this project	BSD-3-Clause

This repository is private until redistribution of the kitchen and apartment scenes is confirmed.