Gemini-PINO Model Checkpoints - Anonymous Submission

Pre-trained Physics-Informed Neural Operator (PINO) checkpoints and Gemma 3 fallback models for hierarchical robotic control. This dataset accompanies the anonymous conference submission "Gemini Robotics-ER 1.5 × Physics-Informed Neural Operators: Toward Robots that Think in Language and React in Physics at Low Latency". Contents: - PINO model checkpoints for press-fit insertion task (100MB) - PINO model checkpoints for peg-in-slot insertion task (100MB) - Gemma 3 2B fallback model for offline operation (500MB) - Model architecture specifications and training metrics - Evaluation results achieving 0.059ms inference latency Performance: 92.1% success rate (press-fit), 95.4% success rate (peg-in-slot), 3.6× improvement over baselines.