Skip to main content

Real-Time SLAM for Indoor Navigation

Duration: 55 min · Level: Intermediate · Module: 4. Perception & Spatial Intelligence · Focus: SLAM, localization, mapping, navigation

Learning objectives

By the end of this lesson you will be able to explain and apply:

  • ORB-SLAM3 (2021)
  • LiDAR SLAM (LIO-SAM, FAST-LIO2)
  • Semantic SLAM
  • Loop closure
  • Dynamic objects

Why this matters​

Simultaneous Localization and Mapping (SLAM) allows G1 to build a map of an environment while tracking its own position within it — without GPS.

Overview​

Simultaneous Localization and Mapping (SLAM) allows G1 to build a map of an environment while tracking its own position within it — without GPS. For indoor hospital or home environments, LiDAR SLAM or visual SLAM provides centimeter-level localization accuracy.

Key concepts​

Key idea

ORB-SLAM3 (2021): state-of-the-art visual SLAM; runs in real-time on CPU; supports monocular, stereo, and RGB-D; robust to dynamic scenes with people moving

  • LiDAR SLAM (LIO-SAM, FAST-LIO2): centimeter accuracy in large environments; too heavy and expensive for lightweight humanoids but used in industrial deployments
  • Semantic SLAM: extends metric SLAM with object labels; "I am 3m from the kitchen counter and 1.5m from the chair" — enables more natural task planning
  • Loop closure: detect when robot returns to a previously visited location; correct accumulated drift in pose estimate; critical for long-duration operation
  • Dynamic objects: humans, doors, moving furniture are obstacles to SLAM; modern approaches use semantic segmentation to mask dynamic elements from map updates
  • Map types: occupancy grid (navigation), point cloud (3D geometry), semantic map (object labels + positions); G1 needs all three updated in real-time

Check your understanding​

Try to recall each answer before expanding it.

Q1. What do you know about ORB-SLAM3 (2021)?

state-of-the-art visual SLAM; runs in real-time on CPU; supports monocular, stereo, and RGB-D; robust to dynamic scenes with people moving

Q2. What do you know about LiDAR SLAM (LIO-SAM, FAST-LIO2)?

centimeter accuracy in large environments; too heavy and expensive for lightweight humanoids but used in industrial deployments

Q3. What do you know about Semantic SLAM?

extends metric SLAM with object labels; "I am 3m from the kitchen counter and 1.5m from the chair" — enables more natural task planning

Q4. What do you know about Loop closure?

detect when robot returns to a previously visited location; correct accumulated drift in pose estimate; critical for long-duration operation

Q5. What do you know about Dynamic objects?

humans, doors, moving furniture are obstacles to SLAM; modern approaches use semantic segmentation to mask dynamic elements from map updates

References​

  • ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM — Campos et al. (2021). IEEE T-RO 2021

← Previous: 4.1 Sensor Suite Design for Humanoids · Next: 4.3 3D Gaussian Splatting for Robot Scene Understanding →

Part of Module 4: Perception & Spatial Intelligence.