Thang To

About

Thang To is a leading researcher in robotic perception and manipulation, with a focus on bridging the sim-to-real gap through synthetic data. His work centers on deep learning for object pose estimation, semantic grasping, and human-robot interaction, enabling robots to reliably perceive and interact with household objects. To’s most influential contribution is his 2018 paper on deep object pose estimation for semantic robotic grasping, which has garnered 283 citations by demonstrating how synthetic data can train deep neural networks for robust manipulation while overcoming the reality gap. He also created the Falling Things (FAT) dataset, a synthetic benchmark for 3D object detection and pose estimation that advances robotics research. To has developed innovative systems for learning human-readable plans from real-world demonstrations using synthetically trained neural networks, and he introduced a method for camera-to-robot pose estimation from a single RGB image, trained entirely in simulation. His work on sim-to-real directional semantic grasping using deep reinforcement learning further exemplifies his commitment to practical, data-efficient robotic systems. Through these contributions, To has significantly advanced the field of robotic perception and manipulation, making synthetic training a viable pathway for real-world deployment.

Research Focus

Key Achievements

Top Papers

1. 1
   Deep Object Pose Estimation for Semantic Robotic Grasping of Household\n Objects

   283 citations · 2018
2. 2
   Falling Things: A Synthetic Dataset for 3D Object Detection and Pose Estimation

   15 citations · 2018
3. 3
   Synthetically Trained Neural Networks for Learning Human-Readable Plans from Real-World Demonstrations

   6 citations · 2018
4. 4
   Camera-to-Robot Pose Estimation from a Single Image

   5 citations · 2020
5. 5
   Toward Sim-to-Real Directional Semantic Grasping

   2 citations · 2020

Key Collaborators

Contact & Links