Racing sandbox
Chaos vehicle physics, track gates, lap timing, wrong-way detection, and Lyra-based gameplay systems form the simulation layer.
Learning on Demand Racing
LoDR
An Unreal Engine racing environment where autonomous drivers train, qualify, and race through reinforcement learning.
System
Built for autonomous racing experiments.
Learning pipeline
Learning Agents and PyTorch drive policy training with observations, actions, and reward shaping tuned for fast iteration.
Online services
Leaderboard, replay, league, and model service infrastructure keeps races measurable and repeatable.
Competition
Train, submit, race, compare.
01
Train policies against gated track progress and collision penalties.
02
Promote candidate models into league-ready NPC drivers.
03
Run deterministic events with replay capture and leaderboard scoring.
04
Study telemetry, tune rewards, and send the next generation back out.
Training
From dashboard to dedicated races.
LoDR connects in-editor experimentation, TensorBoard metrics, model configuration, and online race orchestration into one loop for learning agents that get better on demand.