I’m the Director of Evaluation at 1X Technologies. My goal is to deploy generalist machines that grow from experience and correct their own mistakes. I’m building World Models and large-scale evaluation pipelines towards this mission.

Previously, I worked on robotics, perception, and machine learning as a Senior Software Engineer at Waymo and Everyday Robots / X, The Moonshot Factory (formerly Google[X]).

My research has focused on learning algorithms and representation learning to generalize ML model understanding in robotics, computer vision, and self-driving.

• Trained end-to-end perception models deployed on Waymo cars for rare semenatic event understanding and robust to sensor dropout
• Built RetinaGAN, a sim-to-real domain adaptation approach using perception consistency grounding for door opening and object grasping.
• Worked on the SayCan end-to-end manipulation modeling effort, on sim-to-real transfer and imitation / reinforcement learning training.
• Solved door opening across novel buildings and conference rooms with Task Consistency Loss, Variational Information Bottleneck, and interpretability featured in WIRED
• Invented Population Based Augmentation

its<firstname><lastname> at gmail dotcom