Inspiration
Real-world police driving training is limited by safety, physical space, weather conditions, and the difficulty of recreating unpredictable high-risk scenarios. Events such as sudden pedestrian crossings, vehicle pursuits, or reduced-visibility driving cannot be safely or consistently practised in physical training circuits.
This challenge motivated us to explore how simulation technology could provide a safer, more flexible, and more data-driven training alternative.
What it does
PRISM (Police Response & Incident Simulation Module) is a high-fidelity driving and incident-response simulator designed to replicate complex urban policing scenarios in a controlled virtual environment.
The simulator allows trainees to experience:
Realistic emergency vehicle handling
AI-driven traffic and pedestrian behaviour
Dynamic incidents such as jaywalkers, fallen trees, emergency lane changes, and pursuits
Fully dynamic weather and day–night cycles
Each training run can be replayed and analysed using built-in telemetry tools, enabling objective performance review rather than subjective observation.
How we built it
PRISM was developed using Unity, leveraging its physics system, AI navigation tools, and real-time rendering capabilities.
Key systems implemented include:
A vehicle physics model tuned for emergency driving behaviour, with steering wheel and pedal hardware integration
AI-driven traffic and pedestrian systems, capable of lane-following, yielding, stopping, overtaking, and responding to traffic lights
A dynamic incident system that introduces unpredictable but controllable events during training sessions
A dynamic environment system, supporting weather changes and day–night cycles
Replay and telemetry pipelines that capture speed, braking, steering, and event timelines for post-session analysis
The simulation environment was designed to feel unpredictable while remaining repeatable, ensuring realism without sacrificing training consistency.
Challenges we ran into
One major challenge was balancing realism and control. Traffic and pedestrian AI needed to behave naturally without becoming chaotic or unfair for trainees.
Another challenge involved performance optimization, as large urban scenes with multiple AI agents and dynamic events can quickly impact frame rate. Careful tuning of update cycles, physics interactions, and AI evaluations was required to maintain smooth real-time performance.
Integrating telemetry and replay systems also required careful synchronization to ensure data accuracy across different simulation states.
What we learned
This project reinforced how powerful simulation can be when combined with behavioural AI and data analytics.
We learned that effective training simulations are not just about visual realism — they depend heavily on:
Emergent behaviour rather than scripted outcomes
Measurable performance data
Safe, repeatable exposure to high-risk situations
PRISM demonstrates how simulation can meaningfully support public-safety training, offering insights that are difficult or impossible to obtain through traditional methods.
Log in or sign up for Devpost to join the conversation.