Chapter 29: Localization and Mapping (SLAM)

"An agent becomes interesting at the exact moment perception changes what it dares to do next."

A Patient Embodied AI Agent

Chapter Roadmap

• 29.1 Where am I and what does the world look likelocalization and mapping are coupled because every pose estimate affects the map and every map update affects later pose estimates.
• 29.2 Odometry and dead reckoningodometry integrates motion increments, so small bias compounds into large pose error.
• 29.3 Localization (Monte Carlo / particle filters)particle filters represent pose belief as many weighted hypotheses.
• 29.4 Mapping and occupancy gridsmapping converts sensor rays and poses into a belief over free, occupied, and unknown cells.
• 29.5 SLAM: graph-based and visual SLAMgraph SLAM turns poses and landmarks into constraints, then solves for the trajectory and map that best satisfy them.
• 29.6 Neural and Gaussian-splat SLAMneural and Gaussian representations fold appearance into mapping, offering dense reconstructions for view synthesis and robot inspection.
• 29.7 Map uncertaintyuncertainty is a planning signal, not an afterthought.
• 29.8 Modern SLAM Systems And Failure ModesModern SLAM is no longer one algorithm. It is a contract among inertial sensing, visual or lidar front ends, factor-graph optimization, map maintenance, semantic structure, and failure replay.

# Create a small local environment for the chapter lab.
python -m pip install numpy

# Start a Chapter 29 evidence manifest.
# Add one row per section and keep metrics construct matched.
sections = ['29.1', '29.2', '29.3', '29.4', '29.5', '29.6', '29.7']
manifest = [
    {"section": s, "metric": "closed_loop_success", "perturbation": "occlusion_or_noise"}
    for s in sections
]
print(manifest[0])

# Start a Chapter 29 evidence manifest.
# Add one row per section and keep metrics construct matched.
sections = ['29.1', '29.2', '29.3', '29.4', '29.5', '29.6', '29.7']
manifest = [
    {"section": s, "metric": "closed_loop_success", "perturbation": "occlusion_or_noise"}
    for s in sections
]
print(manifest[0])
for row in manifest:
    row["baseline_score"] = 0.72
    row["shortcut_score"] = 0.81
    row["failure_label"] = "perception_or_planning_interface"
print(manifest)

Use this chapter as a complete teaching unit: concept, minimal implementation, library shortcut, diagnostic perturbation, and postmortem. The pattern prevents a perception model from being evaluated in isolation and never tested as part of the agent loop.