👷 Spatial & Embodied Engineer

👷

Spatial & Embodied Intelligence

Build the system: Geometry → 3D → SLAM → Spatial Memory → Navigation → Robot.

1. 1
   b01What is Spatial Intelligence?45 min
   From cognition to computation: what it means for machines to understand space.
2. 2
   b02Camera and Geometry60 min
   Pinhole cameras, intrinsics/extrinsics, epipolar geometry and triangulation.
3. 3
   b03Depth and Point Clouds60 min
   Stereo, monocular depth, sensors and the most universal explicit 3D representation.
4. 4
   b04NeRF and Gaussian Splatting90 min
   Implicit radiance fields vs explicit Gaussian primitives — the modern 3D stack.
5. 5
   b05Dynamic 3D / 4D Worlds75 min
   Scene flow, dynamic occupancy and 4D representations: add the time axis.
6. 6
   b06State Estimation75 min
   Kalman/EKF/UKF, factor graphs and nonlinear least squares over Lie groups.
7. 7
   b07SLAM and VIO90 min
   Simultaneous localization and mapping; visual-inertial odometry on real benchmarks.
8. 8
   b08Spatial Memory60 min
   Maps, place recognition and persistent scene representations — remembering the world.
9. 9
   b09Affordance45 min
   From "what the world is" to "what the world lets me do".
10. 10
    b10Navigation75 min
    Geometry, semantics and language-guided navigation — spatial intelligence’s killer app.
11. 11
    b11Robot World Models75 min
    Sim-to-real, system identification and world models that survive contact with physics.
12. 12
    b12VLA and World-Action Models75 min
    Vision-Language-Action models and predicting the consequences of actions.
13. 13
    b13Capstone: Build Your Own Spatial World Modelproject
    End-to-end system: perception → state estimation → planning, evaluated on a benchmark.

░░░░░░░░░░░░░░░░░░░░ 0%0/13 completed

Next up: b01 · What is Spatial Intelligence?

Why this order

The track follows the real engineering pipeline of a spatial system:

1. Perception & representation (b01–b05) — geometry first (cameras, depth, point clouds), then the modern neural stack (NeRF, Gaussian Splatting, dynamic 4D).
2. State estimation (b06–b08) — filtering and factor graphs, SLAM/VIO on real benchmarks, and persistent spatial memory.
3. Acting in space (b09–b12) — affordances, navigation, and robot world models up to VLA and world-action models.
4. Capstone (b13) — perception → state estimation → planning, evaluated on a benchmark, end to end.

Exit criteria

You are done with this track when you can stand up a complete spatial-intelligence system — perceive, estimate, plan — and evaluate it on a public benchmark (EuRoC-class). Career narrative: Spatial AI Engineer / Robotics Perception Engineer.

Not your track? The parallel World Model Scientist track builds the models themselves — same foundations, different exit.