AIGP — AI Race Optimizer

Race Performance Analytics

Gate-by-Gate Timing (seconds)

Speed Profile Through Course

Actual Optimal

Lap Comparison

Current Best Theoretical

Corner Efficiency

—

Line Deviation

—

Detection Rate

—

Avg Gate Time

—

AI Model Status

U-Net Segmentation

5ms | 200 FPS | 14.8 MB

YOLOv8n Detection

12ms | 83 FPS | 6.2 MB

Color HSV (VQ1)

0.5ms | 2000 FPS | 0 KB

RL Policy (PPO)

0.2ms | 5000 FPS | 0.4 MB

Pipeline Latency

U-NET 5ms + RANSAC 2ms + PnP 1ms = 8ms

PnP Accuracy

92%

RL Training Progress

0 / 5,000,000 steps

Budget @ 120 Hz

8ms / 8.33ms available

Training Control Center — Cloud GPU Options

Click any card to expand training commands and cost estimates. Prices as of March 2026.

FREE$0 — get started immediately

BUDGET$0.15 – $3.19/hr — best value for training

ENTERPRISE$2 – $30/hr — managed pipelines, multi-GPU

PLATFORMSTracking, hosting, edge inference

Quick Actions

🧠

Train U-Net

~15min A100

👁

Train YOLO

~45min A100

🎮

Train RL

~2hr 1M steps

📦

Export ONNX

U-Net + YOLO

⚡

Benchmark

Vision FPS

🚀

Submit

Package zip

Live AI Race Analyst • Powered by Cloudflare Workers AI

Ask a question about your race performance, configuration, or training strategy. The AI analyst runs on Cloudflare Workers AI (Llama 3.1 8B) at the edge.

Endpoint: aigp-ai.blake-37e.workers.dev/analyze

LLM Integration Options

Claude API

Anthropic • claude-opus-4-6 / claude-sonnet-4-6

Best for code generation, config optimization, race log analysis, failure mode diagnosis

Use: pipe race JSONL logs → Claude analyzes → outputs tuned race_config.yaml

GPT-4o

OpenAI • gpt-4o / o3

Code generation, training data augmentation, multi-modal analysis of FPV frames

Use: analyze FPV screenshots for gate detection failures, suggest HSV ranges

Cloudflare Workers AI

Cloudflare • You have an account

Edge inference, low-latency model serving, global CDN for model weights

Use: serve ONNX gate detector at edge, <10ms inference, $0.01/1K requests

Local LLMs

Llama 3.3 70B • Mistral Large • Qwen 2.5

On-device inference, no API costs, private data. Run via ollama or vLLM

Use: real-time race commentary, on-drone decision augmentation (future)

LLM Use Cases for Racing AI: • Auto-tune PID gains from race JSONL logs • Generate synthetic training data (augmented gate images) • Analyze failure modes (why did the drone miss gate 7?) • Optimize racing line from telemetry path data • Code review: suggest pipeline optimizations