Traffic map
Real corridor data drives deployment precision.
TomTom traffic telemetry identifies congestion hotspots. Census and land-use data predict rider demand. Depot markers show charging infrastructure capacity.
Autonomous Mobility Twin
Autonomous Orchestrator
Real-time corridor telemetry, reinforcement learning policy, and energy-aware fleet scheduling converge in a single control room.
1.35M
Lives Lost Annually
in road crashes worldwide
The Human Cost
94% of crashes are caused by human error (distraction, fatigue, impaired judgment). Autonomous vehicles eliminate these failures with 360° sensors, instant reactions, and perfect attention.
Data Source
WHO Global Status Report on Road Safety 2023
70B+
Hours Driving Annually
wasted in traffic congestion
The Economic Cost
Time lost to congestion costs the EU €270B per year. AVs can optimize routes, reduce bottlenecks, and free up capacity, but cities need data to know where to deploy them.
Data Source
INRIX 2023 Global Traffic Scorecard
€2-5M
Wasted on Studies
that take 3+ years to complete
The Deployment Gap
Cities want AVs but feasibility studies are expensive, slow, and often deploy fleets in the wrong corridors. This platform cuts planning to weeks with data-driven simulations and council-ready reports.
Data Source
McKinsey Urban Mobility Report 2024
The Solution: This Platform
Most cities spend years debating where to deploy AVs, burning fees while missing the window to act. This platform flips that script.
18 months
vs 3-5 years traditional
€840M
estimated savings over 10 years
60%
faster council approval
Control Room
Ingolstadt Autonomous Mobility Twin
Explore traffic patterns, run the RL optimizer, and compare optimized AV performance against INVG's current 274-bus fleet.
Updated ...
Key Features
Three integrated views
Corridor telemetry, optimization engine, and deployment metrics update together. No technical background required.
Policy levers
Tune the optimizer, see instant impact.
Adjust demand weight, charge threshold, and exploration rate. The platform recalculates reliability, energy cost, and grid stress in real time.
Rollout plan
Ingolstadt is the example, not the limit.
Drop in data from Antwerp, Lyon, or Hamburg and the same loop highlights new corridors, depots, and fleet counts overnight.
Preparing city twin…
Blue heat patches visualize inferred rider demand based on where traffic congestion is worst.
How It's Calculated:
• Corridors with high delay (from TomTom traffic) get brighter blue
• Corridors with more route segments show higher density
• Logic: High congestion = more people trying to move = higher ride demand
Limitation: This is a proxy, not real ride booking data. In production, cities would feed actual transit ridership or ride-hailing patterns. For this demo, traffic delay serves as a reasonable demand indicator.
City baseline
These metrics show current system performance. When you run the optimizer, watch how RL improves each one.
Corridors in focus
These are the three corridors with the highest delay right now.
Corridor stats will load once the TomTom feed responds.
RL policy studio
Ten simulated hours run each time the optimizer runs. Each episode adjusts the weights below, then the system keeps the best reward to update the deployment plan.
Epoch0/10
Reliability0.0 %
Served rides as share of total demand
Energy per ride0.0 kWh
Average kWh required per completed trip
Grid relief0.0 pts
Points shaved from grid stress baseline
Reward avg0.0
Average reward per step across the latest episode
Live logstandby
Click Run Optimization to let the policy re-balance supply, energy windows, and equity guardrails. Updates will appear in this log.
Reinforcement Learning Architecture
Policy network pipeline
Proximal Policy Optimization (PPO) processes live telemetry through two hidden layers, then outputs fleet actions: move, charge, or hold.
Inputs
Delay · Demand · Energy · Battery · Time · Weather
Hidden 1
First encoding layer
Hidden 2
Policy representation
Actions
Move · Charge · Hold
Training cadence
The Python backend runs 8,000 PPO timesteps on startup, then fine-tunes for 2,000 steps each time you adjust policy weights (balancing speed with responsiveness).
Real-time inference
Every optimizer tick feeds fresh telemetry through this network, produces a fleet action distribution, and returns metrics to the dashboard in under 300ms.
How it works
From congestion data to deployment plan
The reinforcement learning engine doesn't just show traffic, it solves it. Watch how raw delay data becomes a concrete AV rollout strategy.
1
Traffic builds up
Corridors like Audi Forum ↔ Hauptbahnhof slow to 15-20 km/h during peak hours (30-40 km/h below posted limits).
2
RL learns patterns
The policy analyzes when demand spikes, where delays hurt most, and which zones need more coverage.
3
Deploy smarter
Platform recommends exact AV counts per corridor, charging windows that avoid grid stress, and service hours.
Reading the map above:
- • Yellow/red glowing lines on the Ingolstadt map show congested corridors (thicker = worse delay)
- • Blue heat patches mark high rider demand zones (where people are waiting)
- • Green depot markers show charging hubs the RL policy uses to balance energy costs
- • The traffic cards below reveal segment-level congestion stats from August 2024 TomTom data
Live Traffic
Corridor congestion data
Loading TomTom traffic stats…
Data Sources
Where the numbers come from
All data is real or based on publicly available infrastructure specs. Here's what feeds the twin:
Integrated
TomTom Traffic Stats
Real probe data from millions of connected vehicles (GPS traces, not satellites). August 2024 Ingolstadt dataset with 5,000+ road segments.
Assumed
Depot Charging Capacity
Estimated based on typical urban depot charging capacity (1-2 MW per site). Locations chosen near major transit hubs (Nordbahnhof station, Audi campus, GVZ logistics area). Actual capacity would come from municipal utility data.
Proxy
Demand Modeling
Derived from TomTom traffic delay patterns. Assumption: higher congestion = higher mobility need. This is a proxy visualization; production systems would use actual transit ridership or ride-hailing data.
Real Data Comparison
INVG Today vs Optimized AV Fleet
Ingolstadt's current public transit (INVG) operates 274 buses serving 15.5M annual riders. Here's how an optimized autonomous fleet would compare using real 2019 INVG operational data.
AVs operate 18+ hours/day vs 12-hour bus shifts. Better utilization means fewer vehicles needed for same (or better) service.
Shorter wait times and dynamic routing attract riders who currently drive or walk. INVG's ridership has stagnated since 2011 (49,908 daily riders).
RL optimizer minimizes empty runs and deadheading. Vehicles reposition only when high-value rides are predicted, unlike fixed bus routes that run empty during off-peak.
Main cost savings come from eliminating driver wages (€35-45K/year each). Remaining staff handle fleet monitoring, maintenance coordination, and customer service.
Data Source: INVG 2019 Annual Report
All INVG figures (274 buses, 15.5M annual riders, 6.32M km traveled) are from official Ingolstädter Verkehrsgesellschaft public records. AV scenario projections based on RL optimizer results shown in this platform, scaled to match INVG's service area (255K population across Ingolstadt + 16 surrounding municipalities).
Important Limitation
The AV scenario is a projection, not a deployed system. The 20% ridership increase assumes AVs attract new riders through better service (shorter wait times, dynamic routing). The 30% distance reduction comes from this platform's RL optimizer eliminating empty bus runs. These are estimates, not guarantees.
Understanding the Metrics
What "baseline" means and how RL improves each metric
What is Baseline?
The baseline represents a hypothetical starting point: if Ingolstadt deployed AVs today with no optimization, the system would serve about 75% of ride requests successfully (similar to current bus reliability). This is the benchmark. When the RL optimizer runs, it tests different fleet sizes and routing strategies. If reliability increases to 91.7%, that's a 16.7% improvement, meaning the optimized AV fleet serves more riders than a naive deployment. Note: If reliability drops (shows negative %), the RL policy is exploring a trade-off, sacrificing some reliability to save energy or reduce grid stress. This is intentional. The agent tests whether serving 5% fewer riders but cutting energy costs by 20% delivers better overall value.
How RL Improves It
The RL agent learns which corridors have high demand and positions vehicles proactively. It minimizes wait times by predicting where riders will be, not just reacting to requests. During training, it deliberately tests bad policies to learn what doesn't work.
How baseline was set: Based on typical public transit reliability benchmarks (70-80% for European cities) and standard AV energy consumption estimates from published pilot studies (Waymo, Cruise, EasyMile). Grid stress calibrated to Ingolstadt's transformer capacity.
What is Baseline?
The baseline is the estimated energy per passenger trip if AVs were deployed without optimization. At 0.60 kWh per ride, this represents a naive strategy (charge anytime, route reactively, accept empty runs). The RL optimizer tests when to charge vehicles (off-peak to save money) and how to route them efficiently (minimize deadheading). Typical improvement: 0.48 kWh per ride (20% reduction), which saves €550K annually on a 120-vehicle fleet serving 18M rides.
How RL Improves It
The agent avoids peak electricity pricing by charging at night and coordinates routes to minimize empty deadheading (driving without passengers). It balances energy costs against service reliability.
How baseline was set: Based on typical public transit reliability benchmarks (70-80% for European cities) and standard AV energy consumption estimates from published pilot studies (Waymo, Cruise, EasyMile). Grid stress calibrated to Ingolstadt's transformer capacity.
What is Baseline?
The baseline is the current peak load on the city's electrical grid when all depot chargers run simultaneously. At 38 points, this represents stress on transformers and feeders. The RL optimizer spreads charging across time and depots. If grid stress drops to 28 points, that's 10 points of relief, meaning you avoid expensive grid upgrades ($5M-20M per substation).
How RL Improves It
The agent staggers vehicle charging so not all AVs plug in at 6 PM (peak demand). It uses time-of-use pricing signals and depot capacity constraints to flatten the load curve, reducing infrastructure costs.
How baseline was set: Based on typical public transit reliability benchmarks (70-80% for European cities) and standard AV energy consumption estimates from published pilot studies (Waymo, Cruise, EasyMile). Grid stress calibrated to Ingolstadt's transformer capacity.
What is Baseline?
The baseline is a composite score combining reliability, energy efficiency, and equity. At 50 points, this represents the current system's overall performance. The RL optimizer maximizes this reward by finding the optimal trade-off. A score of 72 means the new policy is 44% better at balancing all three objectives (serve more riders, use less energy, distribute service fairly).
How RL Improves It
The reward function penalizes unserved riders, high energy costs, and unequal service across neighborhoods. The RL agent explores thousands of policy combinations per training epoch to find the strategy that maximizes this weighted score.
How baseline was set: Based on typical public transit reliability benchmarks (70-80% for European cities) and standard AV energy consumption estimates from published pilot studies (Waymo, Cruise, EasyMile). Grid stress calibrated to Ingolstadt's transformer capacity.
The Key Insight
The RL optimizer doesn't just improve one metric. It finds the optimal balance across all four. Sometimes it sacrifices a bit of energy efficiency to serve 10% more riders, or delays charging by an hour to avoid peak grid stress. That's the power of reinforcement learning: it learns trade-offs that humans can't calculate manually.
European rollout vision
From Ingolstadt sandbox to continental service
TestbedNow
Ingolstadt twin is up and streaming.
Four corridors, 120 AVs, and charging load already feed this layout (the exact setup the jury will see first).
ExpansionQ1 2026
Next cities plug in without redesign.
Drop in corridor and depot data from Antwerp, Lyon, or Hamburg and the same workflow highlights their fleet plan overnight.