Jacky Liang

Gemini Embodied Reasoning pre-training, post-training, and foundation models for robotics

Co-Runner and Editor of Skynet Today, publishing AI news without the hype. I write and edit our weekly "Last Week in AI" newsletter, news briefs, editorials, and overviews.

Co-advised by Professors Oliver Kroemer and Maxim Likhachev. Generalizable robot manipulation

Worked on In-Hand Object Pose Tracking via Contact Feedback and GPU-Accelerated Robotic Simulation

Worked on ISAAC and Flex, applying GPU-accelerated physics simulation to robotics and robot learning.

- Work under Professor Ken Goldberg for Robotics and Robot Learning. - Dex-Net - Studying the application of AI and machine learning in the autonomous learning of robust robotic grasps. - Deep Imitation Learning - Applying deep learning techniques to learn robot tasks from high dimensional images. - Developed pipelines for feature generation of grasps and predicting grasp success metrics both in simulation and in physical experiments using Scikit-Learn and TensorFlow. - Work used to build Dex-Net 2.0, a 99% precision grasp planning system using CNNs to predict robust grasp qualities from depth images. See http://berkeleyautomation.github.io/dex-net/ - Built DY-Teleop, a real-time teleoperation system to facilitate data collection for deep learning from demonstrations for bilateral manipulation tasks. Best Student Paper Award Finalist at CASE 2017. See https://berkeleyautomation.github.io/teleop/ - Worked with Intuitive Surgical’s da Vinci Research Kit and the ABB YuMi. Released open source python - YuMi interface yumipy. See https://github.com/BerkeleyAutomation/yumipy - Implemented Multi-Armed Bandits interface for physical experiments to sample and learn grasps through empirical trials. - Developed numerous utilities in grasp generation, pruning, and analysis, including collision detection, mesh subdivision, and safety protocols for multi-threaded serial communications with robots. - Designed tools to analyze statistics and error modalities of robot controllers.

Taught: - CS188 Introduction to Artificial Intelligence. - EE16A Design of Information Devices and Systems with topics in Linear Algebra and Circuit Design.

Built tool to monitor data aggregation systems for GSuite applications to improve system reliability and reduce maintenance overhead.

-Built internal SQL database tool to perform schema analysis, data table health checks, and generate stored procedures. Alleviated the need for manual schema reviews. -Developed intelligent log message grouping and diffing interface for dealing with large volumes of exceptions coming live from server farms to help devs better understand and fix bugs. -Created automation tool to sync and create endpoints that communicate with Amazon’s SNS and SQS. -Designed, tested, and shipped new product features used by millions of users. -Technologies used: .NET MVC, TFS, AWS, TypeScript, AngularJS