SAW-Bench: Learning Situated Awareness in the Real World

SAW-Bench (Situated Awareness in the Real World) is a benchmark for evaluating observer-centric situated awareness in multimodal foundation models — the ability to reason about space, motion, and possible actions relative to one's own egocentric viewpoint as it evolves over time.

Unlike prior benchmarks that emphasize environment-centric relations (how objects relate to each other in a scene), SAW-Bench probes whether a model can maintain a coherent observer-centric spatial state from egocentric video. It comprises 786 real-world videos captured with Ray-Ban Meta (Gen 2) smart glasses and 2,071 human-annotated question–answer pairs across six tasks. Even the best model trails humans by 37.66%.

• 📄 Paper: arXiv:2602.16682
• 🌐 Project page: https://sawbench.github.io/
• 💻 Evaluation code: https://github.com/UCSB-AI/SAW-Bench

Tasks

Task (task)	What it probes	# QA
localization	Where am I within the space (corner / side / center)?	200
direction	Where is a target relative to my current heading?	834
shape	What is the shape of the path I traveled?	546
revplan	How do I get back to where I started?	229
memory	What changed in the scene between two visits?	100
affordance	What action is feasible from my current pose?	162
Total		2,071

Dataset structure

ucsbai/SAW-Bench/
├── data/                       # one Parquet shard per task + all.parquet
│   ├── localization.parquet · direction.parquet · revplan.parquet
│   ├── shape.parquet · affordance.parquet · memory.parquet
│   └── all.parquet             # all tasks combined
└── videos_compressed/          # egocentric clips
    └── Scene_<id>/<key>.mp4

Fields

field	type	description
task	string	one of the six task names
id	int	id within the task
question	string	the multiple-choice question
options	list[str]	answer choices
ground_truth	string	the correct option text
answer	int	index of ground_truth within options
key	string	"<scene>_<video>", maps to videos_compressed/Scene_<scene>/<key>.mp4
scene_category	string	indoor or outdoor

Usage

Load the QA pairs:

from datasets import load_dataset

# one split per task
ds = load_dataset("ucsbai/SAW-Bench", split="direction")
print(ds[0])

# or all tasks combined
ds_all = load_dataset("ucsbai/SAW-Bench", "all", split="test")

Download the videos:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="ucsbai/SAW-Bench",
    repo_type="dataset",
    allow_patterns=["videos_compressed/**"],
    local_dir="SAW-Bench",
)

For the full evaluation pipeline (download → generate → parse → score), see the code repository.

Ethics, privacy & responsible use

SAW-Bench consists of real-world egocentric videos. Please read this statement before using the data.

Collection & consent. Videos were self-recorded by participants who consented to wearing the camera (Ray-Ban Meta Gen 2 smart glasses) in everyday indoor and outdoor environments. Incidental third parties (e.g., passers-by) and identifiable locations may appear in the background; no individuals were deliberately targeted, tracked, or directed.

Privacy minimization. Audio is removed from all clips. A face/identity-blurred variant of the videos was produced during the study. Even so, faces, license plates, or other identifying details may remain partially visible in some frames.

Permitted use. Non-commercial academic research only, under CC BY-NC 4.0.

Prohibited use. Do not attempt to identify, re-identify, locate, or contact any individual in the videos; do not use the data to train or evaluate face-recognition, biometric, surveillance, or person-tracking systems; do not use it commercially.

Removal requests. If you appear in a video, or are a rights holder, and want a clip removed, contact chuhan_li@ucsb.edu and we will promptly remove it.

By downloading the data you agree to these terms.

Citation

@inproceedings{li2026sawbench,
  title     = {{SAW}-Bench: Learning Situated Awareness in the Real World},
  author    = {Chuhan Li and Rilyn R. Han and Joy Hsu and Yongyuan Liang and
               Rajiv Dhawan and Jiajun Wu and Ming-Hsuan Yang and Xin Eric Wang},
  booktitle = {Forty-third International Conference on Machine Learning},
  year      = {2026},
  url       = {https://openreview.net/forum?id=8lwrYjv6r7}
}