RLAI-1.4M

Introduction

RLAI is a computer vision model that takes an input feed of a Rocket League livestream and generates controller commands to drive the car autonomously. The model takes real-time screen captures and produces both binary (button) and analog (joystick/trigger) outputs that map to a virtual game controller.

Key features:

• Real-time screen-to-controller mapping
• Dual-head architecture for binary and analog controls
• Built for Rocket League gameplay
• Lightweight (1.4M parameters, 5.42 MB)
• Optimized for low-resource environments

Model Details

Architecture Specifications

• Type: Custom CNN (RocketNet) with dual output heads
• Parameters: 1,355,475 (1.4M)
• Model Size: 5.42 MB (FP32)
• Input Resolution: 270x480x3 (BRG)
• Output Dimensions: 19 (11 binary + 8 analog controls)

See github repo for full details: https://github.com/Mockapapella/RLAI