Joseph A. Vincent
Vaughn College of Aeronautics and Technology, Stanford University
Papers
4
Total Citations
44
H-Index
2
About
Joseph A. Vincent is a roboticist whose work sits at the intersection of distributed machine learning, safety verification, and trustworthy autonomy. His research focuses on enabling teams of robots to learn collaboratively under real-world constraints, and on providing rigorous guarantees for the neural-network-driven policies that control them. Vincent’s most influential work, “DiNNO: Distributed Neural Network Optimization for Multi-Robot Collaborative Learning” (2022, 37 citations), introduced a novel algorithm that allows robots to jointly optimize deep neural network models over mesh networks without sharing raw data—a critical capability for scalable, privacy-preserving multi-robot systems. He has also made foundational contributions to safety verification with “Reachable Polyhedral Marching (RPM)” (2021), a method for computing exact reachable sets of ReLU neural networks, enabling rigorous safety analysis of perception and control pipelines. More recently, Vincent has advanced trustworthy evaluation of learned policies, developing statistical frameworks to bound expected performance, value at risk, and generalization from limited real-world rollouts. His work directly addresses the gap between theoretical guarantees and practical deployment, making him a key voice in the push toward verifiably safe and robust autonomous systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4