Robot System Architecture Overview
A robotic system is a complex integration of hardware and software subsystems working in concert. Understanding the architecture helps engineers design, debug, and optimize robot behavior. The architecture defines how sensing, processing, decision-making, and actuation are organized and interconnected.
Primary Subsystems:
- Sensing/Perception: Gathering information about the robot and environment
- Cognition/Processing: Interpreting sensor data, planning actions
- Actuation/Execution: Physically executing planned motions
- Communication: Interfacing with operators, other robots, and infrastructure
- Power: Powering all subsystems reliably
Key Sensors in Robotic Systems
| Sensor Type | Measurement | Technology | Application | Range/Accuracy |
|---|---|---|---|---|
| Encoder | Joint angle/position | Optical/Magnetic | Joint feedback | < 0.001° resolution |
| IMU | Acceleration, rotation | MEMS | Navigation, balance | < 0.1° drift/hr |
| Camera (RGB) | 2D visual data | CMOS | Object detection, vision | 1-12 MP, 30-120 fps |
| Depth Camera | 3D point cloud | Structured light/ToF | 3D perception | 0.3-8m, ±2mm |
| Lidar | 3D environment scan | Laser ranging | SLAM, obstacle avoidance | 0.1-200m, ±2cm |
| Force/Torque | Contact forces | Strain gauge | Manipulation, assembly | < 0.1N resolution |
| GPS/RTK | Global position | Satellite signal | Outdoor navigation | ±2cm (RTK) |
| Ultrasonic | Proximity | Sound echo | Collision avoidance | 2cm-4m, ±1cm |
Control Architectures
1. Hierarchical (Deliberative) Architecture Classical AI approach: Sense → Plan → Act. A top-level planner receives world state, computes optimal plan, sends commands down through layers. Clean and predictable but slow for dynamic environments. Used in early space rovers and factory robots.
2. Reactive (Behavior-Based) Architecture Rodney Brooks' Subsumption Architecture (1986): No central world model. Multiple layers of simple behaviors run in parallel; higher layers can inhibit/subsume lower ones. Fast response to environment changes. Used in early autonomous vehicles and Roomba.
3. Hybrid Architecture Combines deliberative planning for high-level goals with reactive behaviors for low-level obstacle avoidance. Most modern robots use this approach. Example: autonomous cars plan routes using maps (deliberative) while reacting to pedestrians in real-time (reactive).
4. Three-Layer Architecture (Common Standard)
- Deliberative Layer: Long-term planning, task sequencing, goal management
- Executive Layer: Translates plans into action sequences, monitors execution
- Reactive Layer: Real-time sensor-motor loops, safety behaviors
5. ROS (Robot Operating System) ROS is the de-facto middleware for robotics. It provides a node-based communication framework (topics, services, actions), hardware abstraction, device drivers, and a rich ecosystem of packages. ROS 2 (2017+) adds real-time support and improved security.
Levels of Robot Autonomy (SAE-inspired framework)
| Level | Name | Human Role | Robot Role | Example |
|---|---|---|---|---|
| 0 | No Autonomy | Full control, robot is a tool | Execute exact commands | Teleoperated surgical robot |
| 1 | Driver Assistance | Monitor, correct robot | Assist with specific subtasks | Industrial robot with safety override |
| 2 | Partial Automation | Supervise, handle exceptions | Automate specific tasks | Warehouse sorting robot with human oversight |
| 3 | Conditional Automation | Available for takeover when asked | Full task execution in defined conditions | Self-driving car on highway |
| 4 | High Automation | Optional, can disengage | Handle all scenarios in operational domain | Autonomous taxi in geo-fenced area |
| 5 | Full Autonomy | No intervention needed | Fully autonomous in all conditions | Mars rover (delayed communication) |
Key Architectural Components
- 01
World Model / Environment Map: Internal representation of the robot's surroundings — occupancy grids, semantic maps, scene graphs
- 02
Localization Module: Determining the robot's position using SLAM, GPS, or visual odometry
- 03
Path Planner: Computing collision-free paths from start to goal — A*, Dijkstra, RRT, PRM algorithms
- 04
Task Planner: High-level sequencing of subtasks to achieve goals — STRIPS, HTN planning, PDDL
- 05
Trajectory Generator: Converting discrete paths to smooth, physically executable trajectories with velocity profiles
- 06
Controller: Closed-loop feedback control — PID, computed torque, impedance control
- 07
Safety Monitor: Real-time collision detection, force limiting, emergency stop — safety-critical real-time systems
- 08
Communication Interface: ROS topics/services, OPC-UA for industrial systems, WebSocket for cloud connectivity