01/10
01
Aerial view of derby arena

Agents of Destruction

The world's first live autonomous vehicle demolition derby

02
Cockpit perspective

The Concept

AI Against Itself

The world's first autonomous vehicle destruction derby. No drivers. No mercy. Just pure algorithmic chaos in the desert.

Self-driving. Self-destructing.
03

The Core Idea

Adversarial Engineering As Performance

We've trained machines to navigate cities safely. Now we place them inside controlled chaos and watch how intelligence behaves under pressure.

Research Objectives

  • 01 Validate real-time decision-making under multi-agent adversarial scenarios with intentional impact dynamics
  • 02 Capture sensor degradation and system behavior during high-impact collisions—data impossible to ethically generate otherwise
  • 03 Generate training datasets of extreme edge cases for reinforcement learning models

Why This Matters

Current AV testing focuses on known scenarios. Real-world deployment demands resilience against unknown unknowns.

Traditional crash testing validates passive safety. We're validating active decision-making under duress.

Hundreds of millions of simulation miles still miss critical edge cases. Adversarial testing generates the extreme failure modes that matter most.

04

Technical Approach

Build Methodology

Phase 1: Acquisition

Base Vehicles

  • •4x used sedans/vans
  • •Structural reinforcement
  • •Roll cage installation
  • •Drivetrain validation

Phase 2: Autonomy

System Integration

  • •Camera-based vision
  • •GPU computing platform
  • •Real-time path planning
  • •Adversarial behavior models

Phase 3: Validation

Testing & Safety

  • •Simulation testing
  • •Field trials in controlled arena
  • •Remote kill switch validation
  • •Sensor degradation analysis

System Architecture

Perception

  • •8x camera array
  • •360° coverage
  • •Object detection
  • •Trajectory prediction

Compute

  • •NVIDIA RTX GPU
  • •12GB VRAM
  • •100Hz inference
  • •30 FPS processing

Decision

  • •RL agent
  • •Game theory model
  • •Path planning
  • •Risk optimization

Control

  • •Drive-by-wire
  • •CAN bus
  • •5ms response
  • •PID + feedforward

Pipeline →

Sensors → Detection (YOLOv8) → Prediction (3s horizon) → Game Theory → Path Planning (A*) → Control → Actuation

05
Arena competition view

The Format

Different Minds, Different Machines

Each vehicle runs a unique algorithm. Different teams program different strategies. Some prioritize evasion, others aggressive ramming. Competing philosophies of autonomous decision-making in real-time contact.

06
Event atmosphere

The Experience

Spectacle in the Desert

The Broadcast

  • •Live-streamed destruction to global audience
  • •Real-time ML decisions visualized on screen
  • •Predictive betting on collision outcomes

Live Event

  • •In-person spectators at the desert perimeter
  • •Expert commentary breaking down algorithms
  • •Post-match analysis of key decision points

Competitive robotics meets motorsport. Machine learning in the driver's seat. Where cutting-edge research becomes visceral entertainment.

07
Arena aerial view

The Arena

Desert Battleground

Purpose-built 200m × 200m arena in the desert. Reinforced concrete barriers rated for multi-ton impacts. Real-time telemetry broadcast to spectator screens. Autonomous competitors programmed for survival and tactical aggression.

200m
Arena diameter
4
Competing vehicles
08

The Viewing Experience

What You See

Multi-Angle Coverage

  • •Aerial drone tracking the action from above
  • •Onboard cameras showing each vehicle's POV
  • •Perimeter cameras capturing impacts in high-def
  • •Picture-in-picture split screens for simultaneous views

Real-Time Telemetry

  • •Live sensor feed showing what each car "sees"
  • •Decision trees visualizing AI strategy choices
  • •Speed, acceleration, and impact force metrics
  • •Health status bars tracking system degradation

Think Formula 1 telemetry meets battle bots, but every move is calculated by machine learning models making split-second tactical decisions. You can literally see the AI thinking as it dodges, rams, and adapts to damage.

09

Production & Engineering

$300K Budget Breakdown

Vehicle Acquisition & Build

$65,000
Base vehicles & spares$28,000
Structural hardening$28,000
Transport & logistics$9,000

Autonomous Systems

$120,000
Perception & sensors$26,000
Compute platform$24,000
Actuation & control$16,000
Safety systems$8,000
ML development$20,000
Simulation & testing$20,000
Calibration & validation$6,000

Arena & Safety

$35,000
Barriers & perimeter$19,000
Permits, insurance & infrastructure$16,000

Production

$80,000
Camera systems$15,000
Live streaming$7,500
Telemetry visualization$16,000
Control room & post-production$12,500
Media & marketing$29,000
10
Desert scene

The Team

Who's Building This

Hicham Oudghiri

CEO, Enigma

Runs Enigma, an AI company building ML systems to fight financial crime. Spends most of his time designing fraud detection algorithms that help process billions of transactions.

Also makes installations like "The Oracle" at Spring Break Art Fair and "Closets" at Pioneerworks. Likes pushing machines into weird physical spaces.

John Fitzgerald

Artist and Creative Technologist

Makes AI installations that respond to people in real-time with computer vision. Created "The Vivid Unknown" with Godfrey Reggio, a reimagining of Koyaanisqatsi (1982) using generative ML. Premiered at Cannes Film Festival 2025. Was a Visiting Artist at the MIT Center for Art Science and Technology. Recent work premiered at Sundance, Tribeca, BAM, and SIGGRAPH. Interested in making AI feel present rather than programmed.