Curriculum · Level 5 of 5

Autonomous Systems Architect Curriculum

~20 hours study + 45-minute exam 5 modules L4 Systems Engineer + enterprise project lead experience

L5 Autonomous Systems Architect is the top of the BCR technical ladder. You author platform definitions, deploy ML on the edge, navigate the safety standards, and set the direction for how robot families get brought into the service network. This is engineering at the intersection of regulation, machine learning, and mechanical reality.

Modules

1
ISO 10218 — Industrial Robot Safety
3 hours
2
IEC 62061 — Functional Safety
3 hours
3
Edge AI — Jetson AGX Thor & VLA Models
4 hours
4
ML Feature Engineering on Telemetry
4 hours
5
Platform Definition Authoring
4 hours

Module 1 · 3 hours

ISO 10218 — Industrial Robot Safety

ISO 10218 is the international safety standard for industrial robots and robot systems. This module covers Parts 1 and 2, how they apply to humanoids and mobile robots, and what compliance looks like in practice.

Learning Objectives

Summarize Part 1 (robot design) and Part 2 (integration and installation)
Identify the workspace separation and speed limits that apply to a humanoid deployment
Map ISO 10218 requirements onto BCR platform definitions
Know when ISO 10218 does not apply (consumer and research platforms)

Part 1 — design requirements

Part 1 specifies what the robot manufacturer must build in: protective stops, speed monitoring, safety-rated monitored stop functions.

Single-point-of-failure requirements apply to safety-rated functions; standard industrial functions may use standard components.

A humanoid marketed for operation in collaborative spaces must meet the collaborative operation requirements in Part 1.

Part 2 — integration

Part 2 specifies what the integrator must do at installation: risk assessment, safeguarding, verification.

Speed and separation monitoring is the workhorse collaborative mode — the robot slows or stops as a human approaches.

Every installation requires a documented risk assessment. The absence of documentation is itself a finding during audit.

Module 2 · 3 hours

IEC 62061 — Functional Safety

Where ISO 10218 describes what must be true, IEC 62061 describes how to prove it. This module covers Safety Integrity Levels (SIL), how they are determined, and how they shape the safety architecture of an autonomous robot.

Learning Objectives

Explain SIL 1–3 and their probability-of-failure requirements
Perform a risk graph analysis to derive the required SIL for a function
Choose redundant architectures that meet the target SIL
Understand what SIL compliance does and does not guarantee

The SIL levels

SIL 1: probability of dangerous failure per hour between 10⁻⁶ and 10⁻⁵.

SIL 2: between 10⁻⁷ and 10⁻⁶.

SIL 3: between 10⁻⁸ and 10⁻⁷. This is typical for autonomous robots in collaborative operation.

A higher SIL requires redundancy, diagnostic coverage, and independent channels — costs rise quickly above SIL 2.

Module 3 · 4 hours

Edge AI — Jetson AGX Thor & VLA Models

Edge deployment is where ML meets real mechanical constraints: memory bandwidth, latency, power. This module covers the Jetson AGX Thor platform and how to deploy Vision-Language-Action models within its envelope.

Learning Objectives

Know the headline numbers: 275 TOPS, memory bandwidth, TDP
Choose between INT8, FP16, and FP32 for a given latency target
Estimate the memory footprint of a VLA model before deployment
Deploy and profile a real model on device

Thor at a glance

275 TOPS of AI compute, unified memory architecture, full Nvidia CUDA stack.

Memory bandwidth is usually the bottleneck for VLA inference — not TOPS. Profile first, optimize for bandwidth second.

INT8 quantization gives the biggest throughput win, but requires per-model calibration. FP16 is the safe default for new models.

Module 4 · 4 hours

ML Feature Engineering on Telemetry

Raw time-series data is rarely the right input to a classifier. This module teaches the stationary signal features that consistently outperform raw input on vibration, current, and torque classification tasks.

Learning Objectives

Compute RMS, kurtosis, crest factor, and skewness from time-series data
Derive envelope and frequency-domain features
Build a feature pipeline that is consistent between training and inference
Detect concept drift and decide when to retrain

Stationary features that work

RMS: captures overall energy. Rises with imbalance, misalignment, and bearing wear.

Kurtosis: captures 'peakedness'. Rises with impulsive events like bearing spalls.

Crest factor = peak / RMS. Rises as the signal becomes more impulsive.

Skewness: captures asymmetry. Useful for distinguishing bidirectional from unidirectional wear.

Module 5 · 4 hours

Platform Definition Authoring

Bringing a new robot family into the BCR service network means writing its platform definition. This module covers the structure, the required fields, and the review process that gets a new platform into production.

Learning Objectives

Structure a failure-mode taxonomy: component → symptom → root cause
Build the sensor map: what telemetry is available at what rate
Write diagnostic protocols that L1 and L2 technicians can follow
Specify the parts BOM with lead times and supplier alternates

The five required sections

1. Failure-mode taxonomy — every known failure cross-referenced to its symptom and root cause.

2. Sensor map — every telemetry channel, its units, its sample rate, and its diagnostic use.

3. Diagnostic protocols — field-executable procedures for the most common failures.

4. Parts BOM — every replaceable unit, its supplier, and its lead time.

5. Safety and compliance notes — ISO 10218 / IEC 62061 applicability and any platform-specific hazards.

The review process

Draft the definition. At least three real field cases must be back-tested against it.

Internal review by at least two L5 architects. Mechanical, electrical, and ML perspectives must all be represented.

Pilot release to a small group of L3/L4 technicians. Feedback incorporated before general release.

General release. The definition goes into TechMedix and becomes the source of truth for that platform.

References

ISO 10218-1:2025 and ISO 10218-2:2025
Current international standards for industrial robots.
IEC 62061:2021
Functional safety of safety-related electrical control systems.
Nvidia Jetson AGX Thor Developer Guide
Platform reference for the edge compute target.
BCR Platform Definition Template
Authoring template and review checklist for new platforms.

Ready to test?

Take the L5 Exam

5 questions · Passing score varies by level · Results emailed instantly

Back to L5 Start Exam