LLM-Powered HMI Design for FUSO Heavy-Duty Trucks

Chapter 01The FUSO Product & Platform Landscape (2026)

Mitsubishi Fuso Truck and Bus Corporation (MFTBC), a subsidiary of Daimler Truck AG, manufactures commercial vehicles across light-, medium-, and heavy-duty segments primarily for the Japanese and Asian markets. As of March 2026, the HMI designer's product scope spans four distinct vehicle lines with significantly different powertrain, display, and interaction requirements.

Super Great (Heavy-Duty)

The flagship. Powered by the 6R30 12.8L inline-6 diesel with SHIFTPILOT 12-speed AMT. Comprehensive ADAS suite: ABA6 (Active Brake Assist 6), Active Sideguard Assist 2.0, FBIS (Front Blindspot Information System), Active Attention Assist, and LDW. Exhaust aftertreatment: BlueTec SCR + DPF. Telematics: Truckonnect. Right-hand drive for Japan market — critical for zone placement analysis.

eCanter (Light-Duty Electric)

All-electric with modular battery packs (S/M/L configurations). The cluster must handle unique EV states: state-of-charge (SOC), estimated range (eRange), recuperation levels, charging status (AC L1/L2/DC), battery temperature warnings, cell balancing indicators, and V2X readiness. No tachometer — the power/regen gauge replaces it.

Fighter & Canter

Medium- and light-duty diesel platforms. Fighter shares many ADAS systems with Super Great at a lower tier. Canter includes diesel and mild-hybrid variants. Both use simpler cluster configurations but share the core design system with the heavy-duty line.

Chapter 02Regulations, Safety & Display Architecture

Your instrument cluster design must comply with multiple overlapping regulatory layers. Understanding which are mandatory versus advisory is critical for prioritizing design decisions.

Functional Safety — ASIL B Pipeline

The instrument cluster is classified as ASIL B under ISO 26262. This classification defines hard technical constraints on your display pipeline.

Glance-Time Budget

SAE J941's Class B (heavy truck) eye ellipses have not been updated since 2010 and don't account for height-adjustable seats. Treat J941-based cluster placement as approximate.

Display Hardware & CAN Architecture

The instrument cluster receives data from dozens of ECUs via the vehicle's CAN bus network. Understanding this pipeline is essential — it determines what data you can display, how fast it arrives, and what safety constraints apply.

Chapter 03ADAS, Powertrain & Telltale Design

The 2024+ Super Great carries one of the most comprehensive ADAS suites in the Japanese heavy-duty segment. The cluster must handle warning escalation across multiple systems while maintaining clear visual hierarchy.

ABA6 Warning Escalation

Active Sideguard Assist 2.0 monitors blind spots on both sides during turns and lane changes. Three-stage escalation at ≤20 km/h: visual alert → audible alarm + A-pillar lamp → damage mitigation brake. When multiple ADAS systems trigger simultaneously, prioritize by criticality — emergency interventions (ABA full brake, ASA emergency stop) take absolute priority.

Cross-Powertrain HMI Architecture

Diesel (Super Great, Fighter, Canter): DPF regeneration has 4 states: inactive, passive regen, active regen request (pull over), DPF full (derate imminent). AdBlue uses 3-level depletion warning: advisory (>500km), warning (100-500km), critical (<100km — torque derestriction). Both follow ISO 2575 color conventions.

EV (eCanter): The cluster replaces the tachometer with a power/recuperation gauge. Key unique states: SOC with range estimation, charge port status with time-to-full, cell balancing indicator, thermal management warnings, V2X pre-conditioning.

Hydrogen (Hino Profia Z FCV — ARCHION context): Uses 6×70MPa tanks with Toyota MIRAI-derived fuel cell stacks. HMI must display: tank pressure per bank, FC stack temperature, hydrogen leak warning (safety-critical), dual-source range estimation.

Competitive Landscape

Japanese heavy-duty OEM competitors: Hino Profia (now within ARCHION), Isuzu Giga, UD Trucks Quon (Isuzu Motors). European benchmarks: Mercedes-Benz Actros, Volvo FH, Scania S-Series, DAF XG+. None publish detailed display specifications publicly — making LLM-assisted image analysis particularly valuable for building competitor benchmark libraries.

Competitor Benchmark Matrix (Template)

Use an LLM with competitor press photos (see Screenshot Workflows in Chapter 05) to fill this matrix. The empty cells are your research task.

Chapter 04What LLMs Are & How They Work

A Large Language Model (LLM) is a type of AI trained on enormous amounts of text to learn patterns of human language. Given input text, it predicts what should come next. Scale this to billions of parameters and trillions of training words, and next-word prediction becomes surprisingly capable: writing prose, generating code, analyzing documents, and reasoning through multi-step problems.

Think of it as a brilliant, well-read colleague who has never worked at your company, doesn't know your specific situation, and occasionally makes things up with complete confidence. The more context you give, the better the output.

How We Got Here

The Context Window

LLMs process text as tokens — sub-word pieces (~1 token ≈ 0.75 words). Everything you send — your prompt, pasted specs, and the model's response — must fit inside the model's context window. A full DBC file might be 50,000+ tokens. A single ISO standard section might be 10,000 tokens. Current windows range from 4K tokens (GPT-3.5) to 200K tokens (Claude).

Temperature: Controlling Creativity

Temperature controls randomness. Low values produce deterministic output; high values introduce variety. Most models default to ~0.7.

Fundamental Limitations

Chapter 05Models, Tiers & Capabilities

Not all LLMs are equal. Models exist on a spectrum of capability, speed, and cost — and choosing the right tier for the right task is a core skill. Using a flagship model for every quick question wastes time and money; using a small model for complex specification work produces unreliable results.

The Three Tiers

Multimodal Capabilities

Modern flagship and mid-tier models can see images. This unlocks powerful HMI workflows:

• Screenshot analysis: Paste a cluster screenshot and ask "Does this layout satisfy ISO 15005 glance-time constraints?"
• Competitor benchmarking: Upload photos of Actros, Volvo FH, and Scania clusters and ask for a structured comparison of information hierarchy
• Icon evaluation: Share a telltale icon set and ask "Which symbols might be confused at 1.5-second glance time?"
• Handwritten sketches: Photograph a whiteboard sketch and ask the model to convert it into a structured layout specification

Screenshot & Image Workflows

For a UI/UX designer, pasting screenshots directly into the LLM is the single highest-ROI multimodal workflow. You can get design feedback, compliance checks, and competitor analysis from images alone — no specification documents needed.

Model-Specific Behavior

Different model families have distinct personalities. Claude tends toward careful, nuanced responses and will express uncertainty. GPT models tend toward confident, comprehensive outputs. Neither approach is inherently better — but awareness helps you calibrate your verification effort.

Context Window Sizes

The context window determines how much information you can include in a single conversation. For HMI specification work, this matters enormously — a full CAN DBC file, multiple ISO standard excerpts, and your design system documentation can easily exceed 100K tokens.

Chapter 06Prompt Engineering for FUSO HMI

Prompt engineering is the skill of communicating with LLMs effectively. A well-structured prompt is the difference between a vague, generic response and a precise, actionable output tailored to your specific FUSO HMI design task. This is the single most important skill to develop.

The Five Foundational Principles

Putting Principles Into Practice

The five principles above describe what makes a good prompt. Here's the concrete checklist for how to build one:

The Prompt Skeleton

Role:        Who should the AI act as?
Context:     Vehicle, display, CAN signals, standards
Task:        What exactly should it produce?
Format:      Table, JSON, prose, Figma spec?
Constraints: Glance time, ASIL level, RHD, character limits
Examples:    (Optional) Show the pattern you want

Prompt Anatomy — A Real FUSO Example

Here's how the skeleton maps to a real-world prompt:

Six Prompt Patterns for Daily HMI Work

Few-Shot Prompting — Teaching By Example

When zero-shot doesn't produce the format or quality you need, show the model what you want with one or two examples:

Chain-of-Thought — Step-by-Step Reasoning

For complex decisions where you need the model to show its work, ask it to reason step by step. This produces higher-quality analysis because the model can catch its own errors mid-reasoning.

Chapter 07Worked Example: eCanter Charging Status Display

This chapter walks through one complete design task from start to finish — prompt, output, verification, iteration, final spec. Every technique from the preceding chapters is applied in sequence.

The Brief

First Prompt Attempt

Using the prompt skeleton from Chapter 06, we construct a detailed first prompt. This is a realistic first attempt — thorough but not yet accounting for every edge case.

Raw LLM Output

The model returns a structured table. At first glance it looks solid — correct format, reasonable content, professional language. But three errors are embedded in the response.

Verification Pass

Now we apply the verification checklist from Chapter 10. This is the expert's most critical contribution — the model cannot reliably self-verify against domain-specific standards.

Iteration Prompt

We send a targeted follow-up correcting all three errors and adding the missing state. Notice: we don't re-explain the entire context. We reference the prior output and provide specific corrections.

Corrected Output

The model returns a revised specification with all corrections applied.

Devil's Advocate

Before finalizing, we use the devil's advocate technique to stress-test a design decision. We pick one choice that seems reasonable and ask the model to argue against it.

Final Specification

After verification, iteration, and adversarial review, we have a production-ready specification. The devil's advocate pass confirmed the color decision is correct per ISO 2575 but surfaced the fallback scenario, which we address by adding a note about kW display prominence.

Takeaway

Chapter 08Advanced Techniques & Context Engineering

Beyond basic prompting, advanced techniques let you extract consistently higher-quality results, manage long design sessions, and tackle complex multi-part HMI specifications that would be difficult with single prompts.

Self-Consistency Through Multiple Passes

For critical specifications, don't rely on a single model response. Run the same prompt 3 times and compare results. Where all three agree, you have high confidence. Where they diverge, that's exactly where you need human expert judgment. This technique is especially valuable for:

• Warning priority rankings — does the model consistently rank ABA6 above DPF warnings?
• Glance-time estimates — are the timing recommendations consistent?
• Color assignments — does it always map to ISO 2575 correctly?

Reflection & Self-Correction

Ask the model to review its own output before you accept it. This simple technique catches errors the model would miss in a single pass.

Devil's Advocate Technique

Force the model to argue against its own recommendation. This surfaces edge cases and risks you might not have considered.

Context Engineering — Managing Long Sessions

In extended design sessions, the conversation grows and the model's earlier context gets compressed or lost. This is context drift — the model gradually "forgets" constraints you set at the beginning. Symptoms:

• The model stops applying your design system rules
• Outputs start contradicting earlier specifications
• Format consistency degrades
• The model forgets which vehicle model or powertrain you're designing for

The WSCI Framework

Four strategies for managing context effectively — think of them as memory management for LLMs:

System Prompts — Your Persistent Persona

A system prompt sets the model's behavior for the entire conversation. For FUSO HMI work, a well-crafted system prompt eliminates repetitive context-setting.

You are an expert UI/UX designer for Mitsubishi Fuso heavy-duty
truck instrument clusters. You work within these constraints:

Vehicle: Super Great (6R30 diesel, SHIFTPILOT 12-speed AMT)
Market: Japan (right-hand drive)
Standards: ISO 15005 (1.5s glance), ISO 26262 ASIL B (200ms
  telltale latency), ISO 2575 (symbol colors), UNECE R121
Display: Digital cluster, 200ms render budget
Protocol: CAN/CAN FD (J1939), CRC-protected messages

Rules:
- All telltale colors follow ISO 2575 (red/amber/green)
- All text fits within 1.5-second glance budget
- Right-hand drive zone placement (mirror LHD assumptions)
- Flag any specification as UNVERIFIED if based on general
  knowledge rather than confirmed FUSO data
- CAN signal names are illustrative — always note this

Context Window Strategy

Think of your context window as a budget. Here's how to spend it wisely:

Chapter 09Claude Cowork for HMI Design

Claude Cowork (launched January 2026) represents a shift from conversation to autonomous action. Instead of asking Claude questions and receiving text answers, Cowork can independently browse the web, read and write files, execute multi-step research workflows, and iterate on its own results — all while you continue other work.

The Autonomy Slider

Think of AI assistance as a sliding scale, not an on/off switch:

For FUSO HMI design work, the Cowork position is the sweet spot. The model works autonomously on well-defined tasks while you retain oversight and final approval. Full agent mode is for development and testing workflows, not safety-critical specification work.

What Cowork Can and Can't Access

The Agent Loop

When Cowork operates autonomously, it follows a structured loop:

Seven Cowork Use Cases for FUSO HMI

Chapter 10Verification & Safety Protocols

Every LLM output in your HMI workflow must be verified before it enters a specification, gets committed to a design file, or influences a safety-relevant decision. This isn't optional — it's the difference between using LLMs productively and introducing silent errors into safety-critical vehicle systems.

Why Verification Matters More Here

Unlike web design or marketing copy, errors in truck instrument cluster specifications can have physical safety consequences. A wrong color mapping for a brake warning. An incorrect priority ranking that suppresses a critical alert. A telltale timing that violates ASIL B requirements. These aren't cosmetic bugs — they affect driver safety.

The Three Verification Traps

1. Sycophancy — The Model Agrees With You

LLMs have a strong tendency to agree with whatever premise you present. If you say "The DPF warning should be green, right?" the model is likely to say "Yes, that makes sense" — even though ISO 2575 mandates amber for caution states. This is sycophancy, and it's one of the most dangerous failure modes in safety-critical work.

2. Automation Bias — Trusting Because It's AI

When output is well-formatted and confidently stated, humans tend to accept it without scrutiny. A table with perfect markdown formatting, clear headers, and professional language looks authoritative — even if the content contains errors. The more polished the output, the more carefully you need to verify the substance.

3. Consistency Illusion — Same Input, Different Output

The same prompt can produce different results each time due to randomness in token sampling. You might run a compliance check today and get "all clear," then run the identical check tomorrow and get three violations flagged. For critical specifications:

• Run the same verification prompt at least 3 times
• Set temperature to 0 for maximum consistency
• Treat any inconsistency across runs as a flag requiring human expert review

The Verification Checklist

Apply this checklist to every LLM output that will enter a specification document:

Chapter 11ARCHION, Coretura & SDV-Era HMI

The Japanese heavy-duty truck industry is undergoing its largest structural transformation in decades. Two major partnerships will reshape the competitive and technical landscape that your HMI designs must navigate.

ARCHION Corporation (April 2026)

ARCHION Corporation is a holding company that brings Mitsubishi Fuso (Daimler Truck) and Hino Motors (Toyota Motor) under a single corporate umbrella. FUSO and Hino continue to operate as separate brands — this is not a brand merger, but a structural unification creating Japan's largest commercial vehicle group. For HMI designers, the implications are significant: potential design system convergence between FUSO and Hino product lines, shared ADAS platforms requiring consistent warning interfaces, and a combined product portfolio spanning light-duty (eCanter, Dutro) through heavy-duty (Super Great, Profia).

Coretura AB

Coretura AB is a separate joint venture between Daimler Truck and Volvo Group focused on developing a shared Software-Defined Vehicle (SDV) platform. This is purely a software collaboration — the companies remain competitors in vehicle sales. The SDV platform will provide the middleware and software architecture that future truck instrument clusters run on.

What SDV Means for HMI Design

Software-Defined Vehicles separate hardware from software update cycles. For instrument cluster designers, this means:

• Over-the-air updates: HMI changes can ship after vehicle delivery. Your design must handle versioning — different trucks in the same fleet may display different UI versions
• Shared rendering platform: Coretura's middleware may standardize the graphics pipeline across Daimler and Volvo products. Your FUSO-specific design system will need to work within this shared framework
• API-driven displays: Instead of hardcoded screen layouts, clusters may consume display data from a standardized API — making your design specifications more like design tokens and less like pixel-perfect mockups
• Longer design lifecycles: When software updates can change the UI independently of hardware, your initial design must accommodate years of iterative updates while maintaining visual and functional consistency

Hydrogen & Next-Gen Powertrains

The Hino Profia Z FCV (hydrogen fuel cell) joins the ARCHION product family, adding a third powertrain type to your HMI scope. Hydrogen-specific displays require: tank pressure per bank (6x 70MPa), fuel cell stack temperature, hydrogen leak warnings (safety-critical — immediate action required), and dual-source range estimation. Your design system needs to be modular enough to accommodate this without a ground-up redesign.

Chapter 12Expert Workflow & Checklist

You've now covered the complete foundation: the vehicle platforms, the regulations, how LLMs work, how to prompt them effectively, how to use advanced techniques, how to verify their output, and how the industry is evolving. This final chapter puts it all together into a practical daily workflow.

Start Small, Then Compound

Don't try to revolutionize your entire workflow on day one. Start with one use case — perhaps using an LLM to draft alert text for a single warning type. Master that. Then expand to specification tables. Then to compliance checking. Each successful use case builds your prompt engineering intuition and your confidence in verification. This is compounding — small, consistent improvements that accumulate into transformative workflow change.

The Daily Workflow

Common Mistakes to Avoid

Your Expert Checklist

Pin this to your workstation. Before sending any LLM-generated specification forward:

Quick-Reference: Before & After Every Prompt

Measuring Your Impact

Track these metrics to quantify how LLMs improve your workflow:

• Specification draft time: How long from blank page to first complete draft? (Target: 50-70% reduction)
• Compliance review coverage: How many standards clauses checked per review? (Target: 2-3x increase)
• Iteration cycles: How many prompt-adjust-verify loops to reach final spec? (Decreases as prompt skills improve)
• Error rate: Errors caught in verification per specification. (Should decrease as you build better prompts and system contexts)

Glossary

Terms used throughout this reference. Sorted by category.

Appendix: Prompt Library

Every reusable prompt template from this reference, collected for quick access. Replace [BRACKETED TERMS] with your project specifics.

Try This Now — Exercise Prompts