A conceptual training framework exploring how autonomous vehicle support personnel could practice incident classification, escalation, communication, and command decision-making in a simulated environment.
Autonomous vehicle operations create a unique challenge for rider support teams.
When something unexpected happens, an operator may need to quickly understand the situation, communicate with the rider, determine the level of risk, take action within their authority, and recognize when responsibility needs to move to someone with greater decision-making authority.
This project explores how those skills could be developed through scenario-based simulation rather than relying exclusively on classroom instruction or scripted exercises.
The concept combines:
- Dynamic incident scenarios
- Tier-based incident classification
- Defined escalation pathways
- Rider communication practice
- Incident Commander decision-making
- Performance measurement
- After-action review
- Progressive qualification and training records
The objective is not to prescribe how any autonomous vehicle company should operate.
It is an exploration of how structured incident command and competency-based training principles could be applied to autonomous mobility.
Operational incidents rarely arrive with perfect information.
A rider may know something is wrong without understanding why. Vehicle information may indicate one problem while the rider reports another. An operator must determine what matters, communicate clearly, and avoid both underreacting and unnecessarily escalating a manageable situation.
That creates several training challenges:
Recognition
Can the operator identify what is actually happening?
Classification
Can the operator determine the appropriate level of operational concern?
Communication
Can the operator acknowledge the rider, communicate clearly, and establish what happens next?
Authority
Does the operator understand what decisions they can make themselves?
Escalation
Can the operator recognize when responsibility needs to move to another level?
The simulator concept was designed around practicing those decisions together rather than treating them as separate skills.
The prototype uses a five-level incident structure:
| Tier | Classification | Conceptual Response |
|---|---|---|
| 0 | Nominal | System logging and monitoring |
| 1 | Anomaly | Operator acknowledges and investigates |
| 2 | Active Issue | Operator coordinates resolution or field support |
| 3 | Safety Escalation | Incident Commander assumes decision authority |
| 4 | Critical | Emergency coordination may be required |
These tiers are conceptual and were created specifically for this project.
They are not intended to represent the procedures, terminology, response criteria, or escalation structure of any autonomous vehicle company.
The current concept includes four scenario types.
A vehicle stops unexpectedly during a trip and the rider does not understand why.
The operator must investigate the situation, communicate clearly, and determine whether intervention is actually necessary.
A rider reports feeling unwell during a trip.
The scenario tests recognition of a potentially safety-relevant condition and the transition from routine rider support to incident command.
A rider reports an uncomfortable or potentially threatening situation involving another person near the vehicle.
The operator must balance reassurance, information gathering, immediate rider concerns, and escalation.
A vehicle stops before a potentially unsafe roadway condition.
The scenario tests whether the operator can distinguish between a system behaving cautiously and an actual emergency requiring escalation.
The training concept supports progression through increasingly complex operational responsibilities.
Foundation-level scenarios focused on:
- Rider acknowledgment
- Information gathering
- Basic incident recognition
- Tier 1 and Tier 2 response
- Clear communication
More advanced scenarios introduce:
- Safety-relevant conditions
- Escalation decisions
- Authority boundaries
- Structured handoffs
- Coordination with additional resources
The perspective changes from directly managing the rider interaction to directing the operational response.
Training focuses on:
- Decision authority
- Delegation
- Resource coordination
- Escalation management
- Maintaining situational awareness
This creates a progression from performing individual response actions to managing the incident as an operational system.
The prototype explores the use of a the large language model(I used Claude) responsible for generating dynamic scenario interactions and training feedback.
Rather than presenting the trainee with predetermined dialogue, an LLM can allow the simulated rider or operator to react to the trainee's actual decisions and communication.
Conceptually, this allows scenarios to change based on how they are handled.
For example, effective communication may stabilize a rider's concern while poor communication or delayed action may create additional operational complexity.
The prototype also explores using an LLM to support structured after-action feedback.
The interactive prototype is not publicly deployed in this repository.
A production implementation would require appropriate authentication, secure server-side API integration, data controls, testing, and additional engineering safeguards.
This repository therefore documents the operational concept, training methodology, interface design, and intended functionality without publishing API credentials or presenting the prototype as production-ready software.
Scenario completion leads into an After-Action Review intended to evaluate more than whether the trainee selected the "correct" button.
The concept evaluates areas such as:
- Incident classification
- Response timing
- Communication quality
- Escalation decisions
- Authority boundaries
- Closed-loop communication
- Concrete next actions
- Opportunities for improvement
The goal is to turn each simulation into a learning cycle:
Scenario → Decision → Action → Feedback → Improvement
The concept also includes a persistent digital Training Jacket.
Rather than treating simulations as isolated events, performance can become part of a larger competency record.
Potential information includes:
- Scenario history
- Qualification progression
- Communication performance
- Classification accuracy
- Escalation accuracy
- Response-time performance
- AI coaching sessions
- Supervisor evaluations
- Recommended next training
This creates the foundation for connecting simulation performance with a broader qualification and readiness program.
My background is in safety-critical operations, technical training, qualification management, readiness, and operational leadership.
One of the recurring challenges in complex training environments is moving beyond:
"Did someone complete the training?"
and toward:
"Can they recognize the situation, make the right decision, communicate effectively, and perform when the situation changes?"
This project explores how AI-assisted simulation could make that type of training more dynamic, measurable, and repeatable.
The technology is not the objective.
The objective is developing better-prepared operators.
Concept / Interactive Prototype
The project currently explores:
- Scenario-based operational training
- Dynamic AI-generated interactions
- Tier-based incident classification
- Escalation pathways
- RSO and Incident Commander perspectives
- Performance measurement
- AI-assisted coaching
- Digital training records
- Progressive competency development
Future exploration may include:
- Additional incident scenarios
- Multi-vehicle events
- Supervisor dashboards
- Scenario-authoring tools
- Manual evaluator controls
- Organization-level training analytics
- Exportable training records
- Formal qualification criteria
- Learning-management-system integration
My background is in operations rather than software engineering.
I use tools such as AI-assisted prototyping to explore operational problems, test concepts, visualize workflows, and communicate how people, processes, training, and technology might work together.
This repository should therefore be viewed as an operational design and training concept, not as production software or a proposed commercial autonomous vehicle system.
This is an independent portfolio and research project created for educational and conceptual purposes.
It is not affiliated with, endorsed by, or representative of the internal procedures, training programs, safety systems, escalation policies, or operational practices of any autonomous vehicle company or any company actually.
All roles, scenarios, classifications, response structures, workflows, and performance criteria presented here are illustrative concepts developed specifically for this project.
Nicholas Soria