METIS-Core
An open-source deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) framework written entirely in C++ and built on LibTorch. It carries zero Python runtime overhead and is designed for production environments, real-time control systems, and edge device deployment.
Timeline (last 90 days)
METIS-Core is an open-source deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) framework written entirely in C++, which completely removes Python from the training and inference execution loop and runs directly by calling PyTorch's native C++ interface, LibTorch.
METIS-Core was shared on Reddit by developer Felix Romo.
METIS-Core employs a tensor-based abstraction layer where agents handle only pure tensor states, with the intelligence component decoupled from specific application scenarios, enabling adaptation across different domains such as robotics, trading, logistics, and defense.
METIS-Core has built-in support for a Shared-Trunk Multi-Head architecture, where multiple agents share the same neural network trunk to learn joint coordination strategies.
METIS-Core implements a natively running AlphaZero self-play loop, in which Monte Carlo Tree Search (MCTS) is executed in a multi-threaded manner in C++.
METIS-Core integrated a C++ AlphaZero self-play loop in version v0.3.0/v0.3.1-alpha and completed validation on the Los Alamos Chess (6×6 board) benchmark environment.
METIS-Core implemented a multi-head neural network architecture in v0.2.0-alpha and provides a multi-agent cooperative defense reference scenario called SwarmDefenseTIE.
METIS-Core is currently in the alpha stage, is an individually developed project, and benchmark validation is limited to small-scale environments such as 6×6 chess.
METIS-Core's roadmap includes AlphaZero asymmetric MARL (for asymmetric scenarios where agents have different action spaces and objectives) and the addition of Proximal Policy Optimization (PPO) to support continuous action spaces.
All Facts (8)
METIS-Core is an open-source deep reinforcement learning (DRL) and multi-agent reinforcement learning (MARL) framework written entirely in C++, which completely removes Python from the training and inference execution loop and runs directly by calling PyTorch's native C++ interface, LibTorch.
50%UnverifiedMETIS-Core was shared on Reddit by developer Felix Romo.
50%UnverifiedMETIS-Core employs a tensor-based abstraction layer where agents handle only pure tensor states, with the intelligence component decoupled from specific application scenarios, enabling adaptation across different domains such as robotics, trading, logistics, and defense.
50%UnverifiedMETIS-Core has built-in support for a Shared-Trunk Multi-Head architecture, where multiple agents share the same neural network trunk to learn joint coordination strategies.
50%UnverifiedMETIS-Core implements a natively running AlphaZero self-play loop, in which Monte Carlo Tree Search (MCTS) is executed in a multi-threaded manner in C++.
50%UnverifiedMETIS-Core integrated a C++ AlphaZero self-play loop in version v0.3.0/v0.3.1-alpha and completed validation on the Los Alamos Chess (6×6 board) benchmark environment.
50%UnverifiedMETIS-Core implemented a multi-head neural network architecture in v0.2.0-alpha and provides a multi-agent cooperative defense reference scenario called SwarmDefenseTIE.
50%UnverifiedMETIS-Core's roadmap includes AlphaZero asymmetric MARL (for asymmetric scenarios where agents have different action spaces and objectives) and the addition of Proximal Policy Optimization (PPO) to support continuous action spaces.
50%