• Develop learning-based manipulation models for end to end sensor-driven interaction (e.g., reaching, motion generation, and execution in dynamic environments). • Build and maintain manipulation training pipelines: dataset creation from robot logs/teleop, action representations, augmentation, and distributed training. • Design evaluation metrics and regression tests that quantify manipulation reliability, recovery behavior, and safety in real environments. • Develop sim-to-real workflows for manipulation learning, including simulation environments, domain randomization, and failure-mode testing. • Optimize and distill models for edge deployment; benchmark latency, memory use, and stability on target hardware. • Partner with the AI platform team to integrate policies with control and safety systems, and validate end-to-end performance on robots. • Analyze field performance, identify dominant failure modes, and drive iterative improvements through data collection and targeted retraining.