Build the Architecture

Now design the system that satisfies the requirements.

Do not begin with a pile of vendor parts. Begin with responsibilities.

Subsystems

A complete architecture may include:

• mechanical platform;
• power;
• actuators;
• low-level control;
• sensors;
• localization;
• perception;
• mission logic;
• communication;
• operator interface;
• logging;
• safety/stop path.

Not every project needs every block.

Interface diagram

Use boxes and labeled arrows. The first version should show responsibility flow before it tries to look like a finished engineering diagram.

Concept flow

Robot architecture starts with responsibility flow

The diagram is not a parts list. It shows who receives a request, who decides, who measures state, who commands motion, and where evidence is recorded.

1. OPERATORmission request or stop input
   commands
2. MISSION LOGICselects goals and checks authority
   asks
3. LOCALIZATIONestimates where the robot is
   informs
4. MOTION CONTROLturns targets into bounded commands
   drives
5. ACTUATORSmotors and mechanisms change the robot
   moves
6. EVIDENCElogs, pose checkpoints, and measured response

Add power separately. Power and data are not the same arrow.

Trace requirements into architecture

Create a table:

If a requirement has no owner, the architecture is incomplete.

Tradeoff records

Write at least three short architecture decisions.

Example:

ADR-01: Use wheel odometry plus periodic visual correction.
Reason: odometry provides frequent local motion while visual references bound accumulated drift.
Cost: adds camera processing, transform, and visibility dependencies.

Architecture is not the boxes. It is the reasoning behind the boundaries.

Interfaces deserve their own design

Two well-designed subsystems can still fail together if their interface is vague.

For every important arrow in your architecture, define an interface record:

Do this for at least three high-consequence interfaces.

Show power and information separately

A block diagram that mixes power and data into unlabeled arrows hides failure propagation.

Create two views if necessary. The power view answers where energy can fail. The information view answers how state and commands move.

Concept flow

Power moves through a separate dependency chain

Power is not the same arrow as data. A robot can receive the right command and still fail if the energy path cannot support the load.

1. BATTERYstored energy source
   protect
2. PROTECTIONmain breaker, fuse, or safety boundary
   distribute
3. DISTRIBUTIONroutes energy to subsystems
   feeds
4. LOADSmotors, computers, radios, sensors, payloads
   observe
5. EVIDENCEvoltage sag, heat, resets, current, or dropout

Concept flow

Information moves through perception and control

Sensor data becomes estimated state, planning decisions, control outputs, actuator commands, and measured feedback.

1. SENSORSmeasure the robot and environment
   estimate
2. STATEpose, speed, heading, object location
   plan
3. PLANNINGchoose target or next action
   control
4. CONTROLconvert target into safe commands
   command
5. ACTUATORSexecute and create new measurements

Then annotate where the two views meet. A motor driver receives information from control logic and energy from the power system. Either path can fail.

Architecture review question

Take one requirement and trace it end to end.

For a safe-stop requirement, your architecture needs to show the full chain:

Concept flow

A safe-stop requirement has an end-to-end evidence path

The stop input is only the first event. A useful architecture shows authority, command inhibition, actuator response, and measured confirmation.

1. STOP INPUToperator, sensor, or rule requests stop
   evaluated by
2. AUTHORITY LOGICdecides the stop request is valid
   inhibits
3. MOTION COMMANDfuture motion is blocked or limited
   updates
4. CONTROLLER OUTPUTactuator command changes
   measured by
5. CONFIRMATIONmotion stops and evidence is recorded

If your architecture cannot show where the requirement is implemented and how its result is observed, the diagram is not yet useful.

Architecture is not decoration. It is a model you should be able to interrogate.