大学课程
本课程的设计基于对 11 门大学课程的系统调研，覆盖世界模型、3D 视觉、SLAM、机器人学习与空间智能。它们是我们的坐标系：两条路线里的每个模块都会注明所引用的 lecture 和作业。
只链接，不镜像。 下面的卡片全部指向官方课程页面，材料版权归原作者和学校所有。"公开材料"标签表示该课程公开可访问的程度（非常高 → 非常低）。其中两门课，Columbia Spatial AI 与 Harvard GSD Spatial Intelligence，是从建筑与以人为本的设计而不是机器人学的角度切入空间智能，我们把它们标记为互补视角。
公开材料：
University of Pennsylvania
CIS 6280 · World Models
- Probabilistic formulation of world models
- State Space Models & Kalman filtering
- Self-supervised representation learning (JEPA)
- Latent world models, RSSM, Dreamer, TD-MPC
- Diffusion, flow matching & video world models
- 还有 3 个主题
课件讲义作业实验代码视频
💡 本课程的骨干——第一门完全聚焦世界模型的研究生课程。
Stanford University
CS231A · Computer Vision: From 3D Perception to 3D Reconstruction and Beyond
- Camera models & calibration
- Epipolar geometry, stereo, structure from motion
- Optimal estimation (Kalman/EKF/UKF)
- Monocular depth estimation
- NeRF & Gaussian Splatting
课件讲义作业实验代码视频
Carnegie Mellon University
16-825 · Learning for 3D Vision
- 3D representations (explicit, implicit, neural)
- Volume rendering & NeRF
- 3D Gaussian Splatting & differentiable rendering
- Dynamic 3D (Nerfies, scene flow fields)
- Multi-view transformers (VGGT, LVSM)
- 还有 1 个主题
课件讲义作业实验代码视频
Massachusetts Institute of Technology
16.485 · Visual Navigation for Autonomous Vehicles (VNAV)
- 3D geometry & Lie groups (SO(3)/SE(3))
- Two-view geometry, RANSAC, bundle adjustment
- Optimization on manifolds, factor graphs
- Visual-inertial odometry (VIO)
- SLAM formulations & incremental solvers
- 还有 1 个主题
课件讲义作业实验代码视频
📄 MIT OCW: CC BY-NC-SA 4.0 · vnav.mit.edu: CC BY 4.0
ETH Zurich / University of Zurich
151-0632-00L / 03SMDINF2039 · Vision Algorithms for Mobile Robotics
- Perspective projection & camera calibration
- Multiple-view geometry, 8-point algorithm, PnP
- Optical flow & KLT tracking
- Place recognition (bag of words)
- Visual-inertial odometry & sensor fusion
- 还有 1 个主题
课件讲义作业实验代码视频
💡 小项目：在 KITTI/Malaga 真实数据上搭建完整的视觉里程计流水线。
University of California, San Diego
Machine Learning Meets Geometry
- 3D reconstruction (single-image & multiview)
- 3D recognition, detection & segmentation
- 6D pose estimation
- Point cloud & mesh processing
- Part-based generative models
- 还有 1 个主题
课件讲义作业实验代码视频
UC Berkeley
CS294-173 · Learning for 3D Vision
- 3D representations (mesh, point cloud, voxel, implicit/SDF)
- Multi-view stereo & COLMAP
- Differentiable renderers (Soft Rasterizer, Mitsuba 2)
- Implicit differentiable rendering (SRN, DVR, IDR)
- View synthesis (NeRF, MPI, SynSin)
- 还有 1 个主题
课件讲义作业实验代码视频
Cornell University
CS 6672 · 3D Vision
- Image formation, epipolar geometry, SfM
- Localization, 6-DoF pose & SLAM
- NeRF & 3D Gaussian Splatting
- DUSt3R & deep multi-view stereo
- Diffusion & text-to-3D (DreamFusion)
- 还有 1 个主题
课件讲义作业实验代码视频
Technical University of Munich (TUM), Computer Vision Group
Practical Course, 10 ECTS · Deep Learning for Spatial AI
- 3D/4D reconstruction & SLAM (VGGT)
- 3D priors with diffusion models (Bolt3D)
- 3D tracking (SpatialTracker)
- Self-supervised learning with 3D priors (RayZer)
- Bundle adjustment for dynamic scenes
- 还有 1 个主题
课件讲义作业实验代码视频
Columbia University (GSAPP)
ARCH A6956-1 · Spatial AI
🏛️ 建筑 / 设计视角
- Spatial reasoning & spatial ontology
- Generative AI / LLM agents for space
- Vision & spatial language models
- Metric spaces, distance & positioning
- Connectivity, way-finding and way-signalling
- 还有 1 个主题
课件讲义作业实验代码视频
💡 从建筑 / 计算设计视角看空间智能——是补充，不是机器人课程。
Harvard University, Graduate School of Design
SCI-6512 · Spatial Intelligence: Designing the Future of Work
🏛️ 人本视角
- Human-centered spatial intelligence
- Sensors & environmental data collection
- Intelligent environments (ethics & implications)
- Multimodal data analysis & visualization
- Human-space / human-building interaction
- 还有 1 个主题
课件讲义作业实验代码视频
💡 人本 / 智能环境视角；仅课程介绍页公开。