SAIDA RL: An Open-source Reinforcement Learning Platform
summary
An open-source RL platform built to run StarCraft II scenarios with C++ game API, protobuf and shared memory, providing baselines in Python/Keras, a five-step Windows install, visualization and per-scenario metrics. Demonstrations include a micro scenario trained for ~5,000 episodes that learns kiting and corner avoidance, a heterogeneous multi-agent demo where dropship and colliers coordinate, and inverse/imitation learning using 110k pro replays to reproduce distinct play styles. Distributed training support is planned.
AI-written summary · about
chapters (7)
- 0:00Introduction to SAIDA RL
- 3:07Reinforcement learning basics
- 7:53StarCraft AI agent demo
- 11:16SAIDA RL platform overview
- 16:39Agent training demos
- 21:16Research activities and goals
- 26:08Conclusion and Q&A
related talks
37mA practical perspective on training reinforcement learning bots.
24mApplied AI research with StarCraft, Torch and the BEAM
27mSkyRL: A Modular RL Library for LLM Agents
52mReinforcement Learning for Autonomous Cyber Defense
17mRLinf: Flexible and Efficient Large-Scale Reinforcement Learning...
47mAdvanced Reinforcement Learning & Data Science for Social Impact
27mScaling an Open Environments Ecosystem for Reinforcement Learning
47mHow we use Reinforcement Learning (...)