API based reinforcement learning environments.
Sphere provides HTTP+gRPC interfaces for reinforcement learning environments. It grants a common abstraction for RL backends that is accessible from any language.
The API definitions are found in the api package.
There is a test example in test/cartpole, there are many examples in github.com/aunum/gold.
Please open an issue for any features/issues with the project.
To generate the API files
make generate
To build the server images
make build
- Support more language types (s4tf, rust, nim, julia, c)
- Web frontend
- Unity backend
- Multi-agent
- ONNX tensors