Large-Scale Inspection System: Spot

This work presents an advanced inspection system utilizing Boston Dynamics' Spot for autonomous industrial monitoring in collaboration with TSMC. The system integrates Routing System, Multi-Map navigation, and digital twin simulation through NVIDIA Isaac Sim to enhance operational efficiency in large-scale environments. By combining these technologies, we establish a scalable, memory-efficient, and highly reliable framework for autonomous industrial inspection.

Spot Robot

System Framework

The inspection process begins with user-defined task assignments, which are processed by an intelligent routing system that determines optimal navigation paths. The routing system employs:

  • Traveling Salesman Problem (TSP) for single-robot optimization
  • Multi-Agent Path Finding (MAPF) for coordinated multi-robot navigation

Once optimal routes are established, the navigation commands are relayed to the multi-map navigation system, which facilitates seamless movement across different factory sections.

System Framework

Map Switching System

Large factory environments pose challenges in handling comprehensive maps. To address this, we developed a map-switching system utilizing AprilTag markers as positional reference points. This system:

Map Area 1Map Area 2
  • Optimizes memory usage by eliminating the need to load an entire large-scale map
  • Enhances scalability, enabling efficient expansion of mapped areas
  • Ensures precise localization, with AprilTags setting Spot’s initial pose after each transition

By partitioning large spaces into manageable sub-maps and leveraging AprilTag-assisted positioning, we enable reliable and efficient navigation throughout extensive factory environments.

AprilTag Localization

3D LiDAR Localization

Spot employs an onboard 3D LiDAR sensor to continuously scan its surroundings, generating high-resolution point cloud data. This real-time data is cross-referenced with pre-established environmental maps using the 3D Normal Distributions Transform (NDT) algorithm

LiDAR Localization 1
LiDAR Localization 2

System Testing

Outdoor Testing

Indoor Testing

City Science Lab Omniverse Digital Twin

Beyond physical navigation, Spot has been integrated into a digital twin environment via NVIDIA Isaac Sim. This enables:

  • Virtual simulation of navigation scenarios to optimize movement strategies before deployment
  • Real-time visualization of waypoints and routing paths
  • Enhanced predictive analysis, improving operational reliability
LiDAR Localization 1
LiDAR Localization 2
© Copyright 2025 Bruce Lin