SuperMap is a living spatial memory system for embodied AI that fuses high-frequency SLAM with open-vocabulary perception to produce a 4D scene graph for real-time and ROS2-enabled workflows.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
SuperMap is described as a living spatial memory for embodied AI. It fuses high-frequency geometric SLAM with asynchronous open-vocabulary perception to produce 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. It is designed to run online in real-time on robot hardware and offline on datasets, or live via ROS2 from one codebase.
How it works
The project combines SLAM with persistent semantic data to build a 4D scene graph. It supports integration with detectors such as Grounding DINO, YOLOE, and boxer pre-baked detections, and exposes per-frame JSON schema output including fields like bbox3d, label, id, center, spatial_relations, status, and latest_stamp. It provides both offline and live ROS2 modes, subscribing to RGB, CameraInfo, PointCloud2, and Odometry topics in live mode and publishing per-object voxels, labeled boxes, and annotated images.
Getting started
Setup guidance is provided as commands:
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
Options: --detector yoloe|offline, --data_dir <path>, --config <yaml>, --live (rerun window).
Run live ROS2:
# Clone into your workspace src/ as `semantic_mapping`, then:
colcon build --packages-select semantic_mapping && source install/setup.bash
ros2 launch semantic_mapping semantic_mapping.launch.py
In live mode, the system subscribes to RGB, CameraInfo, PointCloud2, and Odometry topics and publishes per-object voxels (/obj_points), labeled boxes (/obj_boxes), and annotated images. Topic, extrinsic, and detector settings are configured in config/semantic_mapping.yaml, and the detection vocabulary is defined in config/prompts.yaml.
Both offline and live modes emit the same per-frame JSON schema (semantic_mapping.serialization) with bbox3d, label, id, center, spatial_relations, status, and latest_stamp, so downstream consumers can use one shared interface.
Recent releases
- None
Traction
- Stars: 57
License
- License: none listed