Skip to main content

Roadmap

Current Version: v0.1.0

OpenRA-RL has completed 6 development sprints with 212+ tests passing.

Completed Features

FeatureStatusSprint
Protobuf schema + gRPC bridgeDone1
C# game engine integration (ExternalBotBridge)Done1
Python environment wrapper (OpenEnv)Done1
Unit tests + integration tests (80 tests)Done2
Live end-to-end verificationDone2
Docker images + Compose deploymentDone3
Null Platform (headless, 3% CPU)Done3
Enriched observations (spatial tensor, unit stats)Done4
21 action types (guard, stance, transport, power)Done4
Real-time bridge (non-blocking, ~25 ticks/sec)Done5
Pre-game planning phase + knowledge toolsDone5
Bulk knowledge tools (faction briefing, map analysis)Done5
Agent fixes (auto-placement, production validation)Done6

Supported Game

PropertyValue
GameRed Alert (OpenRA mod)
MapDefault RA maps
Players1v1 (agent vs built-in AI)
AI DifficultiesEasy, Normal, Hard
FactionsAllied, Soviet (auto-detected)

Upcoming Milestones

v0.2 — OpenRA-Bench + Multi-Agent

  • OpenRA-Bench: HuggingFace Space leaderboard for agent evaluation
    • Standardized evaluation protocol (maps, opponents, metrics)
    • Replay data collection and verification
    • Community submissions
  • Multi-agent support: Agent vs Agent matches
  • Evaluation scripts: Automated N-game benchmarking with metrics export

v0.3 — RL Training Pipelines

  • PPO/SAC integration: Training scripts with TRL and Stable Baselines3
  • Reward shaping: Configurable multi-component reward functions
  • Curriculum learning: Progressive difficulty (Easy → Normal → Hard)
  • Observation encoders: CNN for spatial tensor, Transformer for entity lists

v1.0 — Stable Release

  • Stable API: Versioned Protobuf schema with backwards compatibility
  • Full documentation: API docs, tutorials, research guides
  • Community benchmarks: Published baseline results for common agents
  • Multi-mod support: Beyond Red Alert (Tiberian Dawn, Dune 2000)
  • Replay viewer: Web-based replay visualization for OpenRA-Bench

Contributing

We welcome contributions! Areas where help is needed:

  • Agent implementations: New bot architectures (MCTS, hierarchical RL, self-play)
  • Observation encoders: Neural network architectures for processing game state
  • Documentation: Tutorials, guides, and examples
  • Testing: Edge cases, stress tests, multi-platform verification

See the GitHub repository for open issues.