Street AI — The Industry Operating System for Electric & Autonomous Work Fleets

What Street AI Is

Not fleet management with a chatbot — an operations control layer

Traditional FMS answers "where is the truck, how much fuel, when to service." Street AI answers and executes: which missions must happen today, which vehicle / driver / ADS / upfit fits best, whether energy, credentials, and ODD allow it, what to do when things go wrong — and whether every decision can be traced end to end.

🎯

Unified Object Model

Missions, assets, people, chargers, robots, and sites in one operational data model.

🗺

Real-time Fleet Twin

A digital twin holding the current trusted state of every asset — with source and freshness.

🧭

Industry Mission Engine

Configurable Mission Templates express collection routes, milk runs, work orders, and turnarounds.

🛡

Policy & Safety Plane + Agent Mesh

An auditable agent mesh acting only inside policy, permission, and ODD boundaries.

System Architecture

Street AI overall system architecture

Five layers plus a vertical Trust & Safety plane. Telemetry flows up as trusted facts; missions and signed commands flow down through deterministic services. Watch the data move.

Animated Diagram Street AI Overall System Architecture

Reference: StreetAI System Design & Product Description · Version 1.0 · 2026-07-20. Cloud never holds the last line of safety — on-vehicle ADS and safety controllers complete degradation and minimal-risk maneuvers independently.

Product Layering

Where Street AI sits in the RoboStreet stack

Layer	Position	Responsibility	Boundary
RoboStreet One	Electric robotic truck & work platform	By-wire chassis, low-entry cab, upfits, on-board robots, sensors & compute	Provides controllable, observable, maintainable physical assets
Autonomy Suite	Autonomous driving & machine intelligence	Perception, localization, planning, control, minimal-risk strategy, on-board safety monitor	Owns the dynamic driving task — never driven by generative AI
STREET-CONNECT™	Connectivity, data & fleet base platform	Vehicle connectivity, device onboarding, digital twin, data platform, open interfaces	Supplies trusted facts, command channel, and system of record
Street AI	Industry operations & agent system	Mission orchestration, dispatch, energy, maintenance, safety, compliance, remote ops, optimization	Calls operational tools inside policy boundaries — never replaces vehicle safety controllers

Agent Mesh

Ten specialized agents, one orchestrator, strict permissions

Every automated action must simultaneously satisfy: fresh facts, permission, policy, allowed risk level, and reversibility or approval. Each agent runs the same six-stage runtime: Context Builder → Planner → Deterministic Validators → Risk & Approval Gate → Tool Executor → Evidence Recorder.

Agent	Core goal	Default execution rights
Street Orchestrator	Decompose intent, coordinate agents, form one execution plan	Orchestrates only — never controls vehicles
Mission Planning	Turn demand into missions, stops, resources, constraints	Drafts; release is policy-gated
Dispatch	Dynamic assignment, reassignment, shift recovery	Auto-executes low-risk reversible changes
Energy	Guarantee mission energy; cut peaks and charging cost	Auto-schedules; capacity / tariff changes need approval
Asset Health	Predict failures, create work orders, schedule maintenance	Creates / books; cannot defer safety defects
Safety & ODD Guardian	Validate ODD, geofence, versions, weather, risk	Can block missions; ODD expansion needs approval
Remote Ops Copilot	Summarize anomalies, propose playbooks, coordinate	Authorized workflows only — no high-frequency control
Compliance	Track credentials, HOS/ELD, inspections, permits, deadlines	Reminders & drafts; legal filings need approval
Service Recovery	Handle missed service, delays, congestion, customer comms	Auto-notifies and creates remedial missions per SLA
Sustainability	Compute energy, emissions, utilization, program results	Auto-analyzes; disclosure wording needs approval

Facts fresh∧ Permission∧ Policy pass∧ Risk allowed∧ Reversible / approved ⇒Auto-execution

Unified Operating Loop

Every shift and every driverless mission runs the same eight steps

📥

Demand In

Fixed schedules, work orders, sensor events, 311 requests, WMS/TMS instructions.

🧱

Mission Build

Windows, stops, load, tools, credentials, energy, and ODD compiled into a Mission Plan.

✅

Feasibility Check

Health, software versions, charge margin, weather, geofences, permits, crew, site capacity.

📤

Assign & Release

Optimizer proposes; agent explains; policy engine auto-releases or requests approval.

📡

Execution Monitor

Twin tracks mission, position, SOC, upfit/robot state, risk, and ETA continuously.

🚨

Exception Handling

Playbooks retry, reassign, degrade, remote-assist, service, or go minimal-risk.

🧾

Close & Settle

Evidence, load, charging, defects, hours, service records, and open items reconciled.

📈

Learn & Improve

Plan-vs-actual updates models and templates — production changes still go through approval.

Scale Path

From connected fleet to scaled L4 — gated by evidence, not calendars

0–3 months

Connected Fleet

Assets · telemetry · missions · charging · driver app

3–6 months

Agent-assisted

Dynamic dispatch · energy / health prediction · recovery

6–12 months

Robot-ready

Upfit / robot steps · ODD Registry · event replay · OTA

12–18 months

Supervised Autonomy

Release gates · remote supervision · minimal risk · replay

18–24+ months

Scaled L4

Multi-site · permit packs · capacity & version governance

P95 ≤ 0s

End-to-end live vehicle state

99.95%

Core mission / dispatch API target

Minimum edge offline caching

0k+

Active assets per region (design goal)

Industry Solutions

Four verticals, one core platform

Industry differences live in mission templates, policy packs, forms, and integration adapters — not in code forks.

🏙

Municipal & Waste

Curbside & container collection, sweeping, transfer hauls, storm events, robotic-arm cycles. Integrates GIS, 311/CRM, smart bins, scales, weather.

KPIs: service-point completion, missed-pickup rate, lifts/hour, kWh/route, arm first-pass success, complaint closure time.

📦

Campus & Logistics

Fixed milk runs, closed-loop transport, dock queues, night restocking, robotic load/unload. Integrates WMS/TMS/ERP, access control, RTLS, AMR.

KPIs: on-time mission rate, cycle time, moves/hour, utilization, kWh/cycle, remote-assist resolution time.

🔧

Utilities & Field Service

Inspection, maintenance, storm response, ePTO power, robot/drone carriers. Missions release only when crew credentials, tools, parts, and vehicle capability all pass. OSHA 1910.269 mapped into policy packs.

KPIs: response & first-fix rate, tool readiness, ePTO energy, idle replacement, permit blocks.

✈️

Airports & Ports

Baggage/cargo tugs, turnarounds, yard tractors, gate queues, container moves. Integrates AODB/TOS, gate/OCR, microgrids. FAA AC 150/5210-20A as policy input.

KPIs: moves/hour, turnaround punctuality, dwell time, charger uptime, kWh/move, near-miss rate.

First Pilot

A 90–120 day pilot, designed for evidence

One site, 10–30 vehicles (≥5 electric), 2–3 vehicle types or upfits, one clear industry mission, two shifts, one charging depot, one customer system integration — with 2–4 weeks of baseline data collected before go-live.

Baseline firstCompletion rate, dispatch time, kWh/mission, charge peaks, downtime, missed service

Agents staged inObserve → suggest → low-risk auto-execution, every action auditable

Go / No-Go gatesNo unresolved high-severity safety issues; all automation traceable; measurable improvement without added risk

Weeks 0–2Baseline capture · asset onboarding · charging / OCPP

▼

Weeks 3–6Mission templates · live twin · driver app · reports

▼

Weeks 7–10Dispatch / Energy agents · approvals · service recovery

▼

Weeks 11–17KPI comparison vs baseline · Go / No-Go review