Andy Zeng
Co-Founder & Chief Scientist at Generalist
Palo Alto, California, United States
Summary
🤩
Rockstar🎓
Top SchoolAndy Zeng is a robotics-focused machine learning scientist and entrepreneur with 12 years of experience bridging cutting-edge research and applied systems. As Co-Founder & Chief Scientist at Generalist and former Staff Research Scientist and Tech Lead on DeepMind’s robotics team, he has led work on embodied AI, manipulation, and deep RL that moves from lab prototypes to reproducible code. His open-source contributions include releasing the Ravens robot manipulation code within Google Research and maintaining repositories such as 3DMatch-toolbox and visual-pushing-grasping, reflecting hands-on expertise in 3D perception and learning-based manipulation. Trained at Princeton (PhD) following a double major in Math and CS from Berkeley, he combines rigorous theory with practical implementation and experiment-driven intuition. Colleagues describe him as “one experiment away from magic,” a nod to his iterative, empirical approach to hard robotics problems. Based in Palo Alto, he blends research leadership with active OSS engineering to push robotic capabilities toward real-world deployment.
12 years of coding experience
4 years of employment as a software developer
Bachelors Double Major · Mathematics & Computer Science, Bachelors Double Major · Mathematics & Computer Science at University of California, BerkeleyPhD Computer Science, PhD Computer Science at Princeton University
Github Skills (16)
computer-vision10
pytorch10
machine-learning10
deeplearning-ai10
deep-reinforcement-learning10
artificial-intelligence10
robotics10
deep-learning10
point-cloud10
python10
ai10
reinforcement-learning10
3d10
cprogramming-language9
c-language9
Programming languages (6)
Github contributions (5)
andyzeng/3dmatch-toolbox
Nov 2016 - Apr 2018
3DMatch - a 3D ConvNet-based local geometric descriptor for aligning 3D meshes and point clouds.
Role in this project:
ML Engineer
ML EngineerContributions:54 commits, 25 pushes, 28 comments in 1 year 5 months
Contributions summary:Andy contributed code related to keypoint evaluation and 3DMatch training within the 3DMatch-toolbox repository. They modified the keypoint matching evaluation code, including changes to thresholding and error calculations. The user also added 3DMatch training code specifically for the Marvin framework, indicating involvement in model development and evaluation.
3dpoint-cloud3dmatchrgbdcomputer-vision
andyzeng/visual-pushing-grasping
Apr 2018 - Feb 2020
Train robotic agents to learn to plan pushing and grasping actions for manipulation with deep reinforcement learning.
Role in this project:ML Engineer
ML EngineerContributions:77 commits, 3 PRs, 14 pushes in 1 year 10 months
Contributions summary:Andy appears to be contributing to the development and training of deep learning models for robotic manipulation tasks. Their commits involve modifications to the main training script, including setup options, algorithm configurations, and testing parameters. Code changes reflect adjustments to the model architecture, data loading, and the reinforcement learning training loop. Further contributions extend to debugging and updating the training and model saving procedures.
deep-reinforcement-learningroboticsmanipulationgraspingpushing
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.