AGENTS

End-to-End Closed-Loop Auto Parking (ASPICE Scenario)

Full ASPICE-phased closed-loop AI validation for auto parking, from compliance pre-checks to vehicle demo.

Core Pain Points

1. Complex cross-stage collaboration in automotive software delivery: Compliance pre-check, requirements, architecture, coding, testing, and vehicle demonstration involve multiple stages and roles — easily creating tool and document silos.
2. High cost of traceability and evidence maintenance: Two-way associations need to be maintained across requirements, design, code, tests, and vehicle results; manually maintaining traceability matrices and design evidence is time-consuming and prone to chain breaks.
3. Long integration-validation and on-site demonstration cycle: Software, system testing, and digital-vehicle deployment require multiple rounds of manual preparation.

Solution & Technical Implementation

1. Compliance pre-check: Starts with ISO 26262/21448 and HARA as entry points, first completing safety and compliance constraint checks.
2. ASPICE phased workflow: Sequentially executes system/software requirements analysis, software architecture design, implementation + AUTOSAR configuration, review + UT, and software/system integration testing.
3. Knowledge-base retrieval enhancement: Each stage queries the knowledge base and records design evidence.
4. One-click deployment + on-site demonstration: Deploys validation results to a digital vehicle, executes auto parking, and performs on-site closed-loop demonstration; retains human quality gates at critical stages.

Highlights

• Full-pipeline automation: Connects requirements, code, testing, and vehicle — with humans retained only at key quality gates.
• Whole-chain traceability: Provides a 7-layer two-way traceability matrix, targeting ASPICE 4.0.
• Knowledge-driven + evidence retention: Phase-based knowledge-base retrieval with complete recording of design evidence.
• Runnable and demonstrable: One-click deployment to a digital vehicle validates real closed-loop behavior.

Business Value

• Average end-to-end run time ~90 minutes; average Token consumption ~13M per run, with estimated cost ~USD 225.
• Per-module development cycle ~0.8 person-months vs. 2.0 person-months manually — efficiency improved by 120%.
• Can extend from auto-parking scenarios to cockpit, ADAS, body control, and more automotive-software domains.
• Turns the workflow, traceability matrix, knowledge evidence, and on-site demonstration into repeatable deliverable capabilities.

---