Deepak Pathak

Deepak Pathak’s 2026 AI Engineer conference biography identified him as co-founder and CEO of Skild AI and an assistant professor in Carnegie Mellon University’s Robotics Institute, affiliated with its Machine Learning Department. His robotics work centers on general-purpose intelligence that can learn across different robot bodies, tasks, and environments.

In his presentation, Pathak argues that robotics needs a general brain and scalable, diverse training data. He describes Skild’s “omni-bodied” approach: one model shared across robot hardware, pre-trained using simulation and human videos, post-trained with teleoperation data, and improved through deployment experience fed back into training. He evaluates data by its scalability, diversity, and closeness to the robot’s own joint-angle ground truth.

Pathak presents examples including a gripper inserting imitation AirPods without magnetic assistance, learning from human videos with less than an hour of additional robot data, and making omelets with a roughly $4,000 camera-equipped setup. He explains camera-to-motor control and why climbing unfamiliar stairs requires perception and adaptation beyond the body control needed for a backflip. He also reports GPU assembly work for NVIDIA and package delivery applications, and presents robots adapting to disabled legs or jammed wheels as examples of a model that can accommodate changing bodies.