Papers
6
Total Citations
400
H-Index
4
About
Brian Hou is a leading researcher in robot motion planning and manipulation under uncertainty, with key contributions spanning cloud-based grasp planning, Bayesian reinforcement learning, and autonomous off-road navigation. His seminal work on Dex-Net 1.0 (374 citations) pioneered the use of cloud robotics for robust grasp planning, introducing a Multi-Armed Bandit model with correlated rewards that leveraged a network of 3D objects to dramatically improve grasp success rates. Hou has significantly advanced planning algorithms for problems lacking optimal substructure, developing novel approaches that balance risk and efficiency in environments with hazardous zones. His more recent work on Stein Variational Probabilistic Roadmaps and Bayesian Residual Policy Optimization demonstrates his commitment to integrating probabilistic inference with reinforcement learning for informed decision-making under uncertainty. Notably, his research on dynamic replanning with posterior sampling and multi-sample long-range path planning addresses critical challenges in off-road autonomous driving, enabling robots to navigate previously unobserved environments while continuously processing noisy local observations. Through his innovative fusion of sampling-based planning, Bayesian methods, and learning, Hou continues to push the boundaries of autonomous systems operating in complex, uncertain real-world environments.
Research Focus
Key Achievements
Top Papers
- 1
- 2Efficient Motion Planning for Problems Lacking Optimal Substructure11 citations · 2017
- 3Stein Variational Probabilistic Roadmaps5 citations · 2022
- 4
- 5
- 6Dynamic Replanning with Posterior Sampling2 citations · 2022