I built and maintained the online RoboMaster simulator used by Team Jiao Loong for RoboMaster 2025. It became the main platform for operator practice, internal scrims, decision dataset collection, and RL experiments.

Why we needed it

2025 rule changes made last season’s data mostly obsolete. We needed a place to:

• Run realistic internal matches and Swiss-round scrims
• Record full match states for sentry decision labeling
• Train operators before regionals and nationals
• Host friendly matches with other schools

Core modules

Module	What it does
Scene modeling	2025 arena, robots, rune, and game objects in Unity
Motion control	Mecanum chassis physics and actuator control
Auto-aim	In-sim turret / targeting pipeline for realistic fights
2.5D navigation	Top-down tactical map + path planning overlays
Online sync	Multiplayer rooms over Unity Online Services
UI / referee	Match HUD, HP, timers, and server-side rule checks

Online multiplayer

We upgraded the simulator from LAN-only to public online rooms. The hard part was collision handling under client-side physics: moving control and collision to the server caused too much latency, so we kept client simulation with anti-clipping safeguards (speed limits + auto separation).

It was not perfect, but good enough for data collection and operator training.

In-match view

AI / data generation

The simulator also feeds the decision stack:

• Match logging for neural task labeling (70+ labeled games)
• Bot framework with team-level strategy, per-robot traits, and delayed task switching to avoid chaotic flickering
• Value-map target selection (HP, level gap, position) before the full hierarchical bot stack
• RL environments exported to Gymnasium for PPO training

Role in the larger system

The simulator connects directly to the sentry decision pipeline: collect → label → train → test in Foxglove/Unity → deploy on the real robot.

Team internal project — code not public.

---

← Back to projects