titan-v2
A neural control policy for MicroDuck Arena Tug of War, developed from titan-v1 through further self-play and fine-tuning for Cup competition.
- Standard 61-dimensional observations and 14 joint actions.
- Deterministic ONNX inference at 50 Hz.
- Observation normalization included.
- A single policy for all opponents.
Evaluated locally against the available Cup roster using official Arena rules.
- Downloads last month
- 26