SuperMap

SuperMap: A Living Spatial Memory for Embodied AI RSS 2026

AirLab, Carnegie Mellon University

Announcement: The code will be released after RSS.

News

• 2026-07-15: Project website and teaser video are now live.
• 2026-07-15: Initial README and project overview published.

SuperMap is a living spatial memory for embodied AI. It perceives the world, remembers its evolution, and supports reasoning and action. It is a training-free spatio-temporal SLAM system that builds a persistent semantic world model. It fuses high-frequency geometric SLAM with asynchronous open-vocabulary perception, producing a 4D scene graph: a queryable map carrying spatial and temporal information for every object, enabling visual-language navigation and long-horizon reasoning on real robots.

Highlights

• 👀 Perceive — Stable identities across occlusions and scene changes via 3D-aware instance association.
• 🧠 Remember — persistent object identities capture long-term scene evolution.
• 💡 Reason — a queryable 4D scene graph supports spatial and tempora reasoning.
• 🤖 Act — spatial memory naturally grounds VLN, VLA, and future embodied AI systems.
• 🔌 Model-agnostic — works with Grounding DINO, YOLOE, boxer pre-baked detections.
• ⚡ Fully online — real-time on robot hardware; runs offline on datasets or live via ROS2 from one codebase.
• 🏫 Field-proven — continuous 2-hour deployment across the CMU campus (interactive demo).

Requirements

• Ubuntu 22.04 / 24.04, NVIDIA GPU with ≥ 16 GB VRAM, Python ≥ 3.10
• ROS2 Jazzy workspace for live mode

Setup

git clone https://github.com/superxslam/SuperMap.git
cd SuperMap
conda create -n supermap python=3.11 && conda activate supermap

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124
conda install cuda -c nvidia/label/cuda-12.4.0
export CUDA_HOME=$(dirname $(dirname $(which nvcc)))

pip install -r requirements.txt
python -m spacy download en_core_web_sm

Run (offline)

python examples/prepare_example_dataset.py   # download example dataset (one-time)
python examples/example.py                   # run the mapping pipeline