NEXUS 2026 · A HUB FOR INNOVATION

MoveWise

An RL‑Powered MaaS Super‑App for Behaviorally‑Aware Sustainable Mobility

Integrating QR Tap‑In/Tap‑Out Data Collection, Insurance, Payment, Carpooling & Nudge‑Based Behavior Change — Solving Giuseppe's Mobility Problem

Route AI

Payment

Insurance

Carpool

Nudges

Sajjad Shahali

Kiana Salimi

Ali Vaezi

Team MoveWise - 11 · BEST Politecnico di Torino · March 7–8, 2026

GitHub Repository 🚀 Live Demo

Repo: github.com/aliivaezii/RL-Mobility-Optimizer

Demo: aliivaezii.github.io/RL-Mobility-Optimizer

170 g

CO₂/pkm by Car

14 g

CO₂/pkm by Train

€510

True Car Cost/mo

−80%

CO₂ Reduction

100K+

Target Students

App Experience

MoveWise Super‑App

One‑stop mobility ecosystem — route planning, QR payment, insurance, carpooling, gamification & AI nudges.

🚀 Live Demo — Working & Ready

🏠 Home

🗺 Route Planner

💳 QR Payment

🛡️ Insurance

🚗 Carpooling

📍 Live Map

React FastAPI Deep Q‑Learning PyTorch Chart.js Docker

Architecture

📱 React UI — MoveWise App

↕

⚡ FastAPI Backend

↕

🧠 Deep Q‑Learning RL Engine

↕

📊 QR Data + Behavioral Model

How It Works

1User scans QR at trip start (tap‑in)

2RL agent observes mode, time, cost

3Personalized route + nudge

4Gamification rewards green choices

5Agent improves via continuous learning

Key Features

Smart Route AI: RL‑personalized suggestions

QR Tap‑In/Out: Multi‑modal payment

Insurance Hub: Onboarding funnel

Gamification: Points & leaderboards

Carbon Tracker: Real‑time CO₂ viz

📊 Data Per Trip

Mode · GPS · Travel Time · Price · Time of Day · Weather · Segments

🧠 RL Agent Performance

92%

Route Acceptance Rate

4.7s

Avg Response Time

DQN

Deep Q‑Network Agent

λ=2.25

Loss Aversion Factor

Personalized nudges · Continuous learning · Generalized cost optimization

Problem Analysis

170

gCO₂/pkm Car

14

gCO₂/pkm Train

€510

/mo True Car Cost

€55

/mo PT Cost

🌍 CO₂ Emissions by Mode (g/pkm)

Turin Context

📊 Mode Split — Turin Metro Area

🏙 Turin Metropolitan Context

🎓100K+University students (PoliTo + UniTo)

🚗58%Private car mode share — highest EU peers

🎯18–30Target age: periurban commuters

🏘2.2MMetro population · 312 municipalities

🕐42 minAverage commute — "Inhabiting Time"

⏱ Giuseppe's Travel Time by Phase

Benchmark

🔍 Competitive Landscape MoveWise wins 6/7

Feature	Moovit	Whim	GTT	MoveWise
Route Planning	✓	✓	✓	✓ RL‑AI
Integrated Payment	✗	✓	✓	✓ QR
Insurance	✗	✗	✗	✓
Car‑User Onboarding	✗	✗	✗	✓
Behavior Change	✗	✗	✗	✓ 4‑Layer
Carpooling	✗	✗	✗	✓
Revenue Model	Ads	Sub	Public	E‑comm

💰 Giuseppe's Monthly Cost

True cost incl. fuel, depreciation, insurance, parking

Car: €510/mo MoveWise: €45/mo → saves €5,580/yr

📚 Key References

Thaler & Sunstein (2021) — Nudge: Final Ed. DOI

Kahneman & Tversky (1979) — Prospect Theory DOI

Hensher & Hietanen (2023) — MaaF DOI

Chorus et al. (2008) — Random Regret Min. DOI

Esztergár‑Kiss (2024) — RL for MaaS DOI

Butler et al. (2021) — MaaS Barriers DOI

Concept & Technical Operation

💡 Concept (Basic Idea)

MoveWise is a MaaS super‑app that uses Deep Reinforcement Learning to transform private car users into sustainable commuters — not by telling them what to do, but by making the green choice the easy, fun, and financially smart choice.

Scientific basis: Combines Prospect Theory (λ = 2.25), Hybrid Utility‑Regret framework, and Context‑Dependent Value of Time to model real human decision‑making, not idealized rational agents.

⚙️ Technical Operation (How It Works)

1. User downloads app via insurance/parking service (Trojan Horse onboarding)
2. QR tap‑in/tap‑out collects trip data (mode, GPS, time, cost) — inspired by Denmark's Rejsekort
3. Deep Q‑Network observes user state s_t = (mode history, time, cost sensitivity, weather)
4. RL agent selects optimal action: route suggestion + personalized nudge
5. Reward signal: r_t = CO₂ reduction + user satisfaction + revenue
6. Agent improves via experience replay & continuous online learning

🧠 RL Reward Function

r_t = −[ w₁·GC + w₂·E + w₃·Ψ_behav + w₄·Φ_constr ] + w₅·R_rev

GC = Generalized Cost · E = Emissions · Ψ = Behavioral Model (Prospect Theory) · Φ = Constraints · R = Platform Revenue

🔄 Nudge Selection RL

Nudge*(i,t) = argmax_n∈𝒩 Q̂(s_i^t, n; θ)

A second RL agent learns which nudge works best for each user — personalized, adaptive, not one‑size‑fits‑all.

Structure & How to Use

🏗 Structure / Technologies Used

Frontend: React.js + Vite + Chart.js + Leaflet Maps
Backend: FastAPI (Python) + RESTful API
RL Engine: PyTorch Deep Q‑Network (DQN) with experience replay
Data Pipeline: QR/NFC tap‑in/out → PostgreSQL → RL training
Deployment: Docker + CI/CD + GitHub Pages

📱 How to Use (Giuseppe's Journey)

0

Onboarding: Downloads app for car insurance → parking finder → True Cost Calculator reveals €510/mo real car cost

1

P&R + Train: 40 min · €45/mo · −50% CO₂ · Preserves productivity

2

E‑Scooter + Train: 30 min · €55/mo · −75% CO₂ · Full independence

3

Carpool + Multimodal: 25 min · €50/mo · −80% CO₂ · Peer matching

🎯 4‑Layer Behavior Change Model

Layer 1: Nudges

Default green options, social proof, loss framing, anchoring (€510 vs €55)

Layer 2: Gamification

Points, leaderboards, streaks, carbon budget, badges & milestones

Layer 3: Economic

Insurance discounts, employer packages, dynamic pricing, referral bonuses

Layer 4: Education

True Cost Calculator, impact reports, "Did You Know?" cards

Sustainability & System Design

♻️ Sustainability Aspects

Environmental: −220 kg CO₂/yr per user (commute); −310 kg total. EU target: −55% by 2030
Economic: Self‑sustaining e‑commerce model — no public subsidies needed. Break‑even at 3–4K users
Social: Reduced car dependency for students; safer travel; inclusive multimodal access for all
GDPR: Anonymized data; transparent consent; 90‑day exit notification; IVASS‑compliant insurance

📈 Revenue & Business Model

€600K

Annual Revenue (10K users)

3–4K

Users to Break Even

Revenue streams: Booking commissions (5–15%) · Subscriptions · Insurance · Corporate packages · Data analytics

📐 Functional System Diagram

👤 User Layer — App UI (React + Leaflet + Chart.js)

⬇ QR Data + GPS

⚡ API Layer — FastAPI REST + Auth + Payment Gateway

⬇ State Vector

🧠 RL Engine — DQN (PyTorch) + Reward Function + Replay Buffer

⬇ Optimal Action

🎯 Nudge Engine — Personalized Route + Behavioral Nudge

⬇ Feedback Loop

📊 Data Store — PostgreSQL + Anonymized Analytics + GDPR Vault

📊 Scientific Metrics

12.1×

Car vs Train CO₂ ratio

8.5×

Cost perception gap

80%

Max CO₂ reduction (Phase 3)

Problems Solved & Benefits

🎯 What Problem Does MoveWise Solve?

1. Car dependency traps students in expensive, polluting commutes — 58% drive despite cheaper PT options existing
2. Behavioral inertia: People stick to cars out of habit, not rational choice — loss aversion (λ = 2.25) blocks mode switching
3. No integrated platform combines all transport modes, payment & nudging in one app for Turin
4. Existing MaaS apps don't change behavior — they only inform or bundle; none actively guide users away from cars

📊 Measurable Benefits

45 min

→

25 min

Commute Time

Carpool + train eliminates 44% travel

€510/mo

→

€55/mo

True Cost

Inc fuel+depreciation+insurance+parking

170 g

→

14 g

CO₂/pkm

EEA 2023 verified emission factors

−80%

CO₂ Reduction

−89%

Cost Savings

310 kg

CO₂ Saved/Yr

€4,100

Saved/Yr/User

🌍 System‑Level Impact (10,000 users)

−2,200

Tonnes CO₂/yr

€1.6M

Societal Benefit/yr

−15%

Congestion

🔄 Behavior Change Funnel

Value Creation

🌱

Environmental

310 kg CO₂ saved/user/yr. 2,200 tonnes at 10K users. Reduces noise, accidents & urban congestion. Aligned with EU −55% 2030 target.

💰

Economic

€4,100/yr saved per user. Self‑sustaining e‑commerce model — break‑even at 3–4K users. Zero public subsidies needed.

🤝

Social

Inclusive multimodal access for 100K+ students. Safer routes. Community via carpooling & gamified leaderboards.

🧠

Behavioral

First platform to actively change car‑user behavior via RL‑powered personalized nudging + psychology‑aware adoption.

🔮 Future Scalability

Year 1: Turin students — PoliTo + UniTo (100K+ students)
Year 2: Turin metro periurban corridors — corporate mobility partnerships
Year 3+: Italian cities with car‑dependency (Milan suburbs, Rome, Naples)
Year 5+: EU expansion — any city with peri‑urban car dependency + PT infrastructure

🚀 Possible Developments

• Autonomous vehicle integration — RL agent adapts to AV ride‑hailing options
• Multi‑city RL transfer learning — pre‑trained models adapt to new cities faster
• Employer mobility budgets — B2B packages for corporate commuter management
• Real‑time dynamic pricing — AI‑optimized fares balancing demand & sustainability
• Open API for cities — anonymized mobility data for urban planning decisions

The Big Picture

📊 Social, Environmental & Economic Impact

−2,200t

CO₂/yr (10K users)

€41M

Saved/yr (10K users)

100K+

Students reached

🌿 Environmental: Equivalent to planting 100K trees/yr at scale
💼 Economic: €600K platform revenue + €41M user savings = net positive for society
🏘 Social: Reduced traffic fatalities, improved air quality, inclusive mobility for all

🛡️ Safety & Accessibility

Safety: Fewer cars → fewer accidents. Turin: ~3,200 road accidents/yr; −15% car trips = ~480 fewer incidents
Accessibility: Multimodal routes for mobility‑impaired users. Wheelchair‑friendly routing. Audio navigation support
Data Privacy: GDPR‑compliant, anonymized analytics, 90‑day exit notification, transparent consent

The One‑Sentence Pitch

"MoveWise uses QR ticketing, nudge psychology, and Reinforcement Learning to transform car users into sustainable commuters — making the green choice the easy, fun, and smart choice."

📈 Revenue Projection

€600K

Annual Revenue (10K users)

3–4K

Users to Break Even

NEXUS 2026 · BEST Politecnico di Torino

Team MoveWise · March 7–8, 2026