Mobile Robot Architecture: Structure and Navigation

Mobile robot architecture defines how a robot's hardware, software, and decision-making layers are organized to enable locomotion, perception, and goal-directed behavior across unstructured or dynamic environments. This page covers the structural components of mobile robot systems, the navigation stack that connects sensing to movement, the operational contexts in which these architectures are deployed, and the design boundaries that separate one architectural approach from another. Understanding this architecture is foundational to evaluating robotics platforms across logistics, defense, healthcare, and field robotics sectors — all of which are catalogued in the Robotics Architecture Authority index.

Definition and scope

A mobile robot is any autonomous or semi-autonomous machine capable of locomotion through its environment without a fixed physical tether to its workspace. Mobile robot architecture refers to the full structural organization of the software and hardware systems that enable this mobility — encompassing perception, localization, mapping, path planning, motion control, and task execution.

The scope of mobile robot architecture is defined by 3 primary axes:

1. Mobility substrate — wheeled, legged, tracked, or aerial platforms
2. Autonomy level — teleoperated, supervised autonomous, fully autonomous
3. Operational domain — structured indoor environments, semi-structured warehouses, or unstructured outdoor terrain

The Robot Operating System 2 (ROS 2), maintained by Open Robotics and now stewarded under the ROS 2 Technical Steering Committee, has become the dominant middleware framework for mobile robot software architecture in both research and commercial deployments. ROS 2's node-based publisher-subscriber model, detailed in the ROS 2 architecture improvements reference, directly shapes how most mobile robot perception and control pipelines are built today.

ISO 8373:2021, published by the International Organization for Standardization, provides the normative definitions that distinguish mobile service robots from fixed industrial manipulators — a classification boundary with regulatory significance in healthcare and public-space deployments.

How it works

The functional architecture of a mobile robot follows a layered processing pipeline. The canonical formulation, known as the Sense-Plan-Act pipeline, organizes processing into three discrete phases:

1. Sense — Sensors including LiDAR, RGB-D cameras, IMUs, wheel encoders, and ultrasonic arrays capture environmental and proprioceptive data. Sensor fusion architecture combines outputs from multiple modalities to produce a coherent state estimate.

2. Localize and Map — Simultaneous Localization and Mapping (SLAM) algorithms maintain a probabilistic estimate of the robot's pose within its environment. The SLAM architecture in robotics reference covers the principal algorithmic families — Extended Kalman Filter SLAM, particle filter methods, and graph-based optimization approaches.

3. Plan — The motion planning architecture layer computes collision-free trajectories from the current pose to a goal. Planners operate at 2 levels: global planners (e.g., Dijkstra, A*) for coarse route selection, and local planners (e.g., Dynamic Window Approach, TEB) for real-time obstacle avoidance within a rolling horizon window.

4. Act — The control layer converts planned trajectories into actuator commands. Differential drive, omnidirectional, and Ackermann steering geometries each require distinct kinematic models. PID and model predictive control (MPC) loops close at frequencies typically between 100 Hz and 1,000 Hz, depending on platform dynamics.

The hardware abstraction layer in robotics isolates these software layers from platform-specific hardware differences, enabling algorithm portability across robot models.

Common scenarios

Mobile robot architecture is applied across 4 major operational categories:

Warehouse and logistics automation — Autonomous Mobile Robots (AMRs) in fulfillment centers operate under structured SLAM using pre-built maps, with dynamic re-planning triggered by obstacle detection. The warehouse and logistics robotics architecture reference details fleet coordination patterns, including the centralized traffic management model used by high-density deployments.

Outdoor and field robotics — Agricultural, mining, and military ground robots face GPS-degraded or GPS-denied environments. These platforms rely on multi-modal localization fusing LiDAR odometry, visual odometry, and inertial navigation. The U.S. Defense Advanced Research Projects Agency (DARPA) has funded multiple programs — including the DARPA Robotics Challenge — that stress-tested mobile architectures under degraded conditions.

Surgical and hospital service robots — Platforms navigating clinical environments must satisfy functional safety requirements under IEC 62443 (cybersecurity) and ISO 13482 (safety for personal care robots). The functional safety ISO in robotics reference documents applicable standards.

Humanoid and bipedal mobile platforms — Legged locomotion introduces dynamic balance constraints absent in wheeled systems. Whole-body control architectures must solve contact dynamics at update rates exceeding 500 Hz. Boston Dynamics' Atlas and Agility Robotics' Digit represent the primary publicly documented reference platforms for this category.

Decision boundaries

Selecting a mobile robot architecture requires resolving 5 structural trade-offs:

1. Reactive vs. deliberative architecture — Purely reactive architectures (subsumption-style, behavior-based) offer low latency but limited goal complexity. Deliberative architectures support long-horizon planning but incur computational overhead incompatible with fast dynamic environments.

2. Centralized vs. decentralized robotics — In multi-robot deployments, centralized architectures provide global optimality at the cost of single-point failure risk. Decentralized approaches improve fault tolerance but complicate consistency guarantees.

3. Onboard vs. offloaded computation — Edge computing in robotics retains low-latency control locally, while cloud robotics architecture offloads map storage, fleet coordination, and model inference to remote infrastructure. Latency budgets below 10 ms for safety-critical control loops typically preclude cloud offloading for real-time actuator commands.

4. Map-based vs. mapless navigation — Grid-based occupancy maps provide high-fidelity spatial models but require initialization and maintenance. Mapless deep learning approaches, covered under deep learning perception in robotics, generalize better to novel environments but exhibit reduced interpretability.

5. Hybrid architecture in robotics — Most production mobile robot systems implement a hybrid deliberative-reactive structure: a deliberative layer for mission and route planning, with a reactive safety layer that can preempt planned behaviors within a defined intervention threshold.