Spatial-Intelligence-Handbook

What makes it different

The Spatial Intelligence Handbook is described as the first cross embodiment spatial AI handbook. Rather than another single robot body survey, it places the same class of problem from manipulation, driving, aerial and marine domains side by side for comparison, which the README says no other source does. It covers SLAM, VIO, 3D reconstruction, sensor stacks and deployment traps on one shared layer built from 3D Gaussian Splatting, VGGT and depth foundation models. The pitch is horizontal rather than vertical, and the content is meant to be living rather than a static document that goes stale after six months. The project is written in Python and documented at kensou.mintlify.app, with a Chinese language description in the README.

The Atlas and daily pipeline

The Atlas is presented as a map rather than a list. Every paper the pipeline rates gets a set of five axis coordinates covering problem, representation, sensor, paradigm and time, and those coordinates accumulate daily as new papers are processed. The README shows a paradigm axis breakdown where VLA leads with 94 entries and world model as policy has 31, noting that most breakthrough ratings land there. A daily arXiv pipeline fetches the latest papers, rates them with lightning, wrench or book icons, and writes deeper analyses into the corpus. The badge counts include 329 papers in the atlas and 47 dissections, where each dissection gives representation choice reasoning, key hyperparameters, shape sanity checks and deployment traps.

Structure and licensing

The repository is organized into nine content directories, with foundations as the shared toolbox across all embodiments, crossing as the irreplaceable unique selling point, and embodiments as the application layer. A where to start path offers a thirty minute framework built from 3DGS family, VGGT as a feed forward 3D paradigm shift, a representative VGGT versus drone VIO comparison, and a sensor budget matrix explaining why manipulation avoids LiDAR, drones avoid RGBD and underwater avoids RGB. The license badge shows CC BY 4.0, though the GitHub metadata marks the license as NOASSERTION, so users should confirm the terms in the repository. The project carries 21 stars with no forks or open issues, and the default branch is main.

Editorial conclusion

The Spatial Intelligence Handbook is a CC BY 4.0 cross embodiment spatial AI handbook on the main branch with 21 stars that maintains a daily arXiv pipeline and a five axis Atlas of 329 papers, though its GitHub license metadata shows as NOASSERTION rather than the CC BY 4.0 the badge claims.